Card game system with automatic bet recognition

Abstract

An exemplary embodiment of the invention includes a bet recognition system for use with a card game. A table is provided for playing a card game including a bet location for each player. An image capture device is positioned in proximity to the bet location and configured to capture an image of a player's bet. An image processor is coupled to the image capture device and configured to process the image by locating at least one chip and creating a signal representing the chip, comparing the signal to a plurality of stored signatures, and when a match occurs, generating a signal representing the bet. In one aspect, the image processor is configured to generate an error signal when unable to match a candidate signature with at least one of the stored signatures.

Description

RELATED APPLICATIONS

This application claims priority to U.S. Prov. No. 60/644,208 filed Jan. 14, 2005, incorporated herein by reference.

FIELD

The invention relates to systems for monitoring casino card games and card game players whereas the system includes a means for automatic bet recognition. In particular, the invention provides a technique for acquiring an image of a bet at particular times during the play of a card game, processing the image, and monitoring the bets made by the game players during the play of the game. The system may further include one or more data input or acquisition devices to input other information such as the denomination and quantity of chips comprising the bet, the denomination and quantity of chips comprising the inventory of the game table's chip tray, to identify all positions for which game players are actively engaged in the play of the game and scan and identify the card value, rank and suit and disposition of all cards dealt from the game deck to a game hand, a game player and/or a game dealer engaged in the play of the game, and a means for determining the outcome, win/loss, of each game hand according to the game rules and so forth.

BACKGROUND

Billions of dollars are wagered annually on card games, therefore a means to accurately monitor, record and identify the amount of each bet and to subsequently determine the outcome (win/loss) of each wager according to the game rules is vital to the successful operation of such card games, as is a real time means for monitoring the inventory of the chip tray associated with the game being played. The amounts bet and the wins and the losses of the games and game players can be significant. Unscrupulous players and employees may be inclined to attempt to cheat during play of the game and/or steal chips from the chip tray. Likewise, it is also possible that an unscrupulous dealer may cheat, or act in cooperation with a player who is cheating. Casinos could save significant amounts of money lost to cheaters and dealer theft with a system that could automatically detect the amount of each bet and then determine the outcome of each game hand, the proper win and loss for each game hand played during a game round or playing session, and a means for reconciling the chip inventory of the chip tray at the beginning of a game round with the chip inventory of the chip tray at the end of each game round in accordance with the amounts won or lost by the players, as determined by the automatic bet recognition system, during a game round.

Conventional types of automatic bet recognition techniques are known. For example, a casino may embed radio frequency identification (RFID) tags in their chips, which are then read by an RFID reader proximate to a betting area. While this technique may be advantageous, it may result in erroneous bets since it may not be able to accurately direct the RFID sensing in a limited region of the table. Patents describing this technique include U.S. Pat. Nos. 5,735,742 and 5,651,548. Another type of automatic bet recognition uses a camera in an attempt to image and then determine the amount of each wager. However, conventional systems of this type use cameras mounted relatively distant from the betting area or use a single camera attempting to image the entire game table, which does not provide an accurate and reliable way to determine the bet. Patents describing this technique include U.S. Pat. No. 6,758,751.

SUMMARY

What is needed is a reliable system for acquiring an image of one or more bets associated with a particular seat, position or game player and then processing the image(s) to accurately determine the value of the wager(s). Preferably the chip(s) comprising the bet would be stacked in a vertical position. Such a system could accurately determine the amount of each player's bet(s) and whether or not a bet was won or lost, which would potentially save casinos significant amounts of money. Further, a reliable means for acquiring an image of the game tables chip tray inventory, and accurately calculating the number and denomination of each chip contained in the chip tray, and determining the total value of the chip tray's inventory, in real time, is also needed. The chip tray would preferably be transparent and any chip tray imaging device would be capable of capturing an accurate image of chips placed horizontally in one or more one-half cylinder shaped tubes comprising the chip tray; and whereas the tubes would be slightly angled upward, from the back of the chip tray closest to the dealer, toward the front of the chip tray which would be closest to the players; and whereas the upward angle of each tube would cause the chips to remain flush against one another increasing the accuracy of the chip tray's imaging device. Integrating the automatic bet recognition device and the automatic chip tray imaging device into the system enables the system to generate a real time tabulation of the win/loss of each player, the win/loss of the game table, and to activate an audible or visual alert when a players bet at the end of a game round is inconsistent with the amount that the bet should be, according to the game rules, at the end of a game round; and/or activate an audible or visual alert when the inventory of the chip tray is inconsistent with the amount of chips that should be in the chip tray when the settlement of the bets for the current game round is completed. Further, the system includes a means for the game dealer to enter into the system the amounts of any cash or call bets that would not be identified by the imaging device.

The invention overcomes a number of aforesaid limitations of conventional systems and provides a system and method that can reliably acquire an image of a bet and then process the image to accurately determine the amount wagered. Aspects of the invention can include one or more data input or acquisition devices to input other information such cash or call bets, the denomination and quantity of chips comprising the inventory of the game table's chip tray, debit and credit transactions (fills, chips returned to casino cage, credit issued to and paid by game players), positions or seats occupied by players engaged in the play of the game, cards dealt to each player hand, each players strategy proficiency, the outcome of each game round and so forth.

An exemplary embodiment of the invention includes a bet recognition system for use with a card game. A game table is provided for playing a card game including a bet location for each player. An image capture device is positioned in proximity to the bet location and configured to capture an image of a player's bet. An image processor is coupled to the image capture device and configured to process the image by locating at least one chip and generating a signal representing the chip, comparing the signal to a plurality of stored signatures, and when a match occurs, generating a signal representing the bet. In one aspect, the image processor is configured to generate an error signal when unable to match a candidate signature with at least one of the stored signatures.

An exemplary embodiment of processing an image representing a plurality of chips having a plurality of different denominations comprises the steps of capturing an image representing the plurality of chips, identifying edges of the plurality of chips, segmenting the image into a plurality of individual candidate chips, generating a signature for each of the candidate chips, identifying each of the candidate chips by comparing the signature of each candidate chip to a plurality of stored signatures representing valid chips and associated denominations, and adding denominations associated with each of the valid chips to determine the wager.

An exemplary embodiment of training an image processor to process an image representing a plurality of chips having a plurality of different denominations comprises the steps of capturing a plurality of images representing chips having different denominations, generating a signature for each of the chips having different denominations, and storing the signatures.

The invention provides numerous aspects and advantages to a card game with automatic bet recognition. Advantages of the invention include the ability to identify chips wagered by a player to automatically determine the player's bet for a game hand and the player's win or loss during a playing session. The invention can also track bets during the course of a game to automatically determine the player's betting strategy relative to one or more card count systems programmed into the system.

An exemplary embodiment of the invention includes a chip tray inventory recognition system for use with a card game. A game table is provided for playing a card game including a chip tray to contain the game table's bankroll. An image capture device positioned in proximity to the chip tray and configured to capture an image of the chips comprising the inventory of the chip tray. An image processor coupled to the image capture device and configured to process the image by locating at least one chip and creating a signal representing the chip, comparing the signal to a plurality of stored signatures, when a match occurs, generating a signal representing the bet. In one aspect, the image processor is configured to generate an error signal when unable to match a candidate signature with at least one of the stored signatures.

An exemplary embodiment of processing an image representing a plurality of chips in a chip tray having a plurality of different denominations comprises the steps of capturing an image representing the plurality of chips, identifying edges of the plurality of chips, segmenting the image into a plurality of individual candidate chips, generating a signature for each of the candidate chips, identifying each of the candidate chips by comparing the signature of each candidate chip to a plurality of stored signatures representing valid chips and associated denominations, and adding denominations associated with each of the valid chips to determine the amount of the chip tray inventory.

An exemplary embodiment of training an image processor to process an image representing a plurality of chips in a chip tray having a plurality of different denominations comprises the steps of capturing a plurality of images representing chips having different denominations, generating a signature for each of the chips having different denominations, and storing the signatures.

The invention provides numerous aspects and advantages to a card game with automatic chip tray inventory recognition. Advantages of the invention include the ability to identify the chip tray inventory during real time and provide a running count of the chip tray's inventory and to automatically determine the game tables win or loss during the play of the game; and/or to automatically determine the win or loss of a plurality of game tables coupled to the system during real time.

DRAWINGS

These and other features and advantages will become better understood with reference to the description, claims and drawings.

FIGS. 1A-B depict a gaming table system including peripheral components, a computer, memory and peripheral interfaces according to an embodiment of the invention.

FIG. 2 is a flowchart depicting a method according to an embodiment of the invention.

FIG. 3 is a digital image of two stacks of gaming chips on a table, acquired with a conventional digital camera, at a resolution of 960×1280×3 bytes (24-bit color).

FIGS. 4A-B are exemplary flowcharts showing methods to process digital images of chips according to embodiments of the invention.

FIG. 5 depicts intersecting lines of vertical and horizontal edges detected in the image of FIG. 3 following flowchart 450 step 468.

FIG. 6 depicts a final segmentation of the image of FIG. 3, after flowchart 450 step 472. Grey areas are filled regions identified as possible chip stacks. White lines are borders of the potential chip stacks, as determined by the method.

FIG. 7 depicts images of the three identified potential chip stacks from FIG. 3, after the completion of the method described flowcharts 400 and 450. The first candidate stack is an edge of a soup can and will be eliminated from consideration in subsequent steps of the method.

FIG. 8 is an exemplary flowchart showing an exemplary method to segment an image of a stack of chips into individual chip images.

FIG. 9A depicts an original digital image of an isolated single stack of chips, and the output of intermediate stages of the method shown in FIG. 8. FIG. 9B depicts the results of horizontal edge detection. FIG. 9C shows the most likely interchip boundaries determined by the method.

FIG. 10 depicts images of several individual chips from the stack of FIG. 9, identified in the chip location step.

FIG. 11 is an exemplary flowchart showing a method to locate color bands in the image of an individual gaming chip, and generate a list of band colors and widths for input to a chip identification method.

FIG. 12 depicts an original digital image of a single gaming chip, including some background areas at the edges of the chip, which is the output of an exemplary method. FIG. 12A depicts the original digital image. FIG. 12B shows the edge boundaries of vertically divided regions of the chip edge, as determined by the method. The smooth black line marks the 3× s.d. of the background, a recommended threshold for determining which peaks are significant. FIG. 12C shows the observed colors of the bands of the chip in 12A, after merging adjacent regions and excluding edge regions. FIG. 12D shows the bands after brightness normalization. The four color bands depicted will be input to the neural network classifier of the method.

FIG. 13 depicts a standard signature pattern of the example chip of FIG. 12, as computed in the exemplary method.

FIG. 14 depicts an exemplary architecture for a radial basis function neural network, suitable for matching lists of transformed colors and patterns to those expected for authorized chip patterns. The input layer of the network includes nodes for each entry in the standard signature pattern of any chip. In this example, the chip signatures include the normalized red and green intensities, and the radial width, of up to four color bands. More bands, and the sizes and colors of enclosed symbols, are included in the signature pattern in other embodiments. The RBF layer comprises two-dimensional Gaussian functions, which are centered at the positions of the various standardized colors of authorized chips. Each input band (three node group) is connected to each RBF node. The RBF nodes are activated proportionately to the probability that the color represented by the node is present among the input bands. The output layer is trained to report the probability that the input band colors and widths match the pattern of an authorized chip denomination. Each output node reports the probability that the input chip is a chip of the denomination represented by that node. The value of the output node is the sum of its inputs, each weighted by a weight associated with each line in FIG. 14, added to a “bias” constant, and operated upon by a “transfer function.” Various linear and nonlinear transfer functions are commonly employed. The preferred embodiment used a sigmoid transfer function (tanh or equivalent), which allows the output values to be interpreted as probabilities.

FIG. 15 depicts the radial basis functions of the second layer of a RBF neural network trained to recognize 15 distinct colors. The number of distinct colors recognized in any particular embodiment will depend on the variety of authorized chip patterns used to train the network, as taught in the method. As clarified from FIG. 15, 15 distinct colors are readily distinguished by this method.

FIG. 16 depicts the output of the neural network depicted in FIG. 14, when presented as input with the standard signature patterns of some of the chip images of FIG. 10. Chips 5, 15, and 18, counting from the top of the stack, are chips of authorized denomination $5, and their outputs are shown as blue bars. Chips 19 and 21 in the stack of FIG. 8 are chips of authorized denomination $1000, shown as green bars. Chips 1 and 3 of the stack of FIG. 8 are unauthorized chips, shown as red bars. The neural network correctly classifies all of the chips.

DETAILED DESCRIPTION

The invention is described with reference to specific apparatus and embodiments. Those skilled in the art will recognize that the description is for illustration and to provide the best mode of practicing the invention. As referred to herein, a game constitutes one or more hands of cards.

A. Card Game System

FIG. 1A depicts a gaming table system 100 according to an exemplary embodiment of the invention. The table includes a card area 102 for placement of the cards during play of the game, and a betting area 104 for the player to place a wager. FIG. 1B is an overhead view of the table system 100 showing the player positions 130a-e, the card area 102 and the respective betting areas 104a-e for the player stations.

The dealer station includes a card shoe 106 for storing the cards that the dealer deals to each of the players. The exemplary card shoe includes a card scanner of the type described in U.S. Pat. Nos. 5,362,053; 5,374,061; 5,722,893; 6,039,650; 6,299,536 and 6,582,301 all incorporated herein by reference in their entirety. The card scanner is connected to an interface 108 that provides a signal to a computer 110, which determines the cards dealt to the players.

The dealer station also includes a chip tray 120 where the dealer stores chips for paying out players when they win and for collecting their bets when they lose. In one aspect of the invention, the chip tray is transparent and a camera 122 is positioned under the chip tray to capture images of the chips in the chip tray 120 during the course of the game, including before and after each round. A light can be included under the tray to provide illumination to the chips. The chip tray camera 122 is connected to an interface 128 that provides information regarding the chips stored in the tray to the computer 110. In an exemplary embodiment, the chip tray camera periodically scans the chip tray and communicates images of the contents of the chip tray to the computer, which can then determine the content of the chip trap by performing image processing as described herein. In one aspect, dealers are issued dealer tracking cards and a dealer tracking card (DTC) reader 126 is coupled to the processor and configured to read dealer tracking cards issued by a casino having information regarding the dealers. In this manner, the invention can track the contents of the chip tray during each dealer's shift and ensure that the wins and losses are correctly paid into and out of the chip tray.

The player station 130 can include an exemplary display to provide player information including betting and game information, casino information and other information via interface 132. A player tracking card (PTC) reader 134 is provided for the player to enter a card and provide betting information to the casino computer for gambling credits and so forth. PTC readers are known in the art and include magnetic, optical and radio frequency identification type systems. A camera 140 is provided in the player station to capture an image of the player's wager in the betting area 104. Also, since a casino may have inconsistent lighting, in one aspect, a light 144 is provided to illuminate the wager in betting area 104. The light provides a consistent image illumination that improves image capture and image processing in some circumstances.

An optional separate camera 148 can also be coupled to system. When the system determines that one or more predetermined criteria relative to the play of the game or the game player has been achieved the camera 148 is programmed to automatically focus in on the play area or seat occupied by the game player and photograph the subject player. The player's photo will be automatically transmitted and stored within a commercially available biometric database maintained by the host casino and activate an audible or visible alert for the benefit of the dealer and or management. The predetermined criteria may include, without limitation, a VIP player, a known card cheat, a known card counter or a player who has been barred from play has logged in as playing at the specific game table in a specific seat, or a player who the system has determined, during the play of the game, to be a highly skilled card counter, is varying his/her bets according to the decks running or true count, winning an unusual amount of money or hands and so forth.

The computer 110 includes a memory 112 that stores information including control procedures 112a, communication procedures 112b and data 112c. The computer 100 is also coupled to a casino computer 150, which collects information regarding the bets and keeps track of the money in the casino. Operation of the computer is described below with reference to the card game system 100.

FIG. 2 is a flowchart 200 depicting a method according to an embodiment of the invention. In step 202, the game begins and the invention performs wager tracking. The invention provides a means for signaling the beginning of a game. In one aspect, the dealer presses a button on the card shoe 106, chip tray 120 or other location to signal the start of the game to the computer. Camera 140 acquires a first image of the bet (initial wager) in the area 104 and the computer stores the image. In one aspect, the computer also processes the image as described below. In step 204, the game is played. In step 206, the game is finished and the camera acquires a second image of the bet (final wager). In step 208, the first image and second image are compared to one another. If the player wins and the bets match, then the player is paid in step 212. If the house wins and the bets match, then the house keeps the wager in step 214. In either event, if step 208 determines that the final wager does not match the initial wager, then step 216 generates an alarm signal to alert the dealer and/or additional casino personnel to review the game for possible fraud. In aspects of the invention, the alarm can be an audible or visual alert in proximity to the game table and/or an alert on a remote monitor in a remote location.

In one aspect, the first image is taken at the beginning of a game and the second image is taken at the end of the game. In other aspects, images may be taken at specific times during the course of play or at periodic time intervals during the course of play. One benefit to acquiring multiple images is that there may be times during the game that a change in the bet is allowed, for example in blackjack, including doubling down or splitting. In these circumstances, the bet may change during the game and additional images acquired during the game may be processed to ensure the lawful play of the game. In one aspect, FIG. 2 includes the additional step 207 to depict intermediate image acquisitions of the player's station during course of the game, and step 207A to depict intermediate image acquisition of the dealer's chip tray during course of the game.

In one aspect, a first image and second image are compared to one another to ensure that no illegal change was made to the bet during the course of the game. One technique for comparison is to permit a certain number of pixel differences between the images that define an allowed match, where a number of different pixels above that threshold number is defined as a mismatch.

In yet another aspect, steps 202A and 220 are implemented to acquire an image of the chip tray and to identify the contents of the chip tray. This can be done in terms of contents or value of all the chips in the tray. Once the game is complete and the invention knows the value of the wagers won and lost by the players, the invention can calculate an expected value of the chips that should be in the chip tray. Step 220 compares the actual amount to the expected amount. If there's a mismatch between these amounts, the invention can alert the dealer or supervisor to investigate.

B. Automatic Bet Recognition Image Processing

An exemplary method of performing an automatic bet recognition technique is described below with reference to the figures. Headings 1 through 8 describe exemplary aspects of the technique for the sake of describing the invention in an organized manner and are not intended to be limiting.

1. Obtain Suitable Digital Image

FIG. 3 depicts chips placed in a betting area similar to area 104. A preferred input to the automatic bet recognition system (ABRS) image processing and pattern recognition (IPPR) system is a medium or high resolution color digital image of the betting area 104. Color resolution should include at least 4 bits of resolution, and preferably at least 8 bits of resolution, in each color channel. The spatial resolution can be comparable to that achieved by commercially available CCD or CMOS imaging devices commonly used in digital cameras (e.g. 1 mega-pixel image or higher). The imaging device should be aimed and focused such that all stacks of chips are viewed from the side and in focus. A stack of chips will therefore appear in the image as a rectangle with vertical and horizontal sides, whose width is constant and determined by the standard diameter of a gaming chip (and the distance between the stack and the imaging device), and whose height varies depending on the number of chips in the stack.

In some embodiments of the invention, the value of chips in a chip tray is to be imaged and evaluated. In these embodiments, the stacks of chips will be horizontal rather than vertical, and will be confined to a constrained area (i.e. the chip tray). In the following discussion, references to “vertical” and “horizontal” dimensions of chip stacks should be understood to be reversed in such embodiments. Also, certain functions, e.g., function 2 describing location of stacks of chips in the image, may be obviated in such embodiments.

All stacks of chips to be evaluated are preferably completely contained within the image, and any extraneous objects (e.g. cards, water bottles, etc.) in the image should be spatially disjoint from the stacks of chips. If more than one stack of chips is present in the betting area, each such stack should be disjoint from and not occlude any other such stacks in the image. The lighting should be as uniform as possible, without distinct shadows superimposed on the stack. The background should contrast as much as possible with the colors of the chips. Such contrast is facilitated if the imaging device is focused at the distance of the betting area, so that distant backgrounds, e.g. players' clothing, is somewhat out of focus.

The exemplary card gaming system is shown with reference to the exemplary embodiment shown in FIG. 1. Details of the exemplary physical embodiment of the imaging hardware used to obtain the required images, its spatial orientation on the game table, and the external signals and/or internal timers used to trigger the acquisition of an image, are components of the ABRS system which employs the method taught in this invention, and may be modified with respect to the ABRS-IPPR. Whenever such an image is obtained, the ABRS-IPPR will proceed to process and evaluate the stacks of chips (if any) within the betting area, using the method taught in this invention. The ABR-IPPR may be used with other configurations than shown in the exemplary embodiment of FIG. 1.

2. Locate Stacks of Chips in the Image

Once a suitable digital image has been obtained and presented to the computer 110 executing the ABRS-IPPR software program, the program first locates the stacks of chips (if any).

Initially, it is beneficial to perform some pre-processing with various digital filters, such as a median filter in order to normalize signal intensity, improve contrast, remove image features much smaller than gaming chips, and filter out other noise.

The invention employs a method of digital image processing. One such method is called “edge detection” and searches for vertical and horizontal lines bordering regions of different color. Another method called “line continuity search” extends horizontal and vertical lines to delimit regions. A complementary method called “regional continuity search” seeks to find regions of consistent color and/or texture within the image, which are candidates for the edge surfaces of individual chips. An alternative method called “template matching” seeks to locate areas of the image that correspond to a preset template, which in this case is a rectangle of known width and variable height and color. The latter method may be performed on the original image, or, preferentially, on a two-dimensional Fourier transform of the image. While these techniques are described with reference to color, they also work well on a grayscale copy of the original color image.

FIG. 4A depicts a high-level method employed by the invention to identify a player's bet, locate the candidate chips constituting the bet, match signatures of the candidate chips to those stored in memory and then validate a match, if possible. Although many algorithmic methods and combinations of algorithmic methods could be employed to locate stacks of chips in an image, the preferred embodiment comprises eleven sequential steps shown in FIG. 4B.

Step 452—Vertical edge detection: Each pixel in the image is replaced by a grayscale value computed from the average difference between the color values of a number of pixels to its left and the color values of a number of pixels to its right. The appropriate number of pixels to use for edge detection depends on the optical and digital resolution of the image; it ranges between one pixel and about five pixels. For the example in FIG. 3, three pixels were used for vertical edge detection.

Step 454—Vertical median filters: Each pixel in the vertical edge image after step 452 is replaced by the median of a vertical column of N pixels, of which it is the center. The effect of this filter is to eliminate vertical lines shorter than N/2, and to fill in gaps in longer vertical lines. Appropriate values of N depend on the image resolution and on the size of vertical boundaries to be detected—values of N from 3 to ½ of the chip height may be optimal. For the example in FIG. 3, in which the average height of a single chip image is 30 pixels, 9 pixels were used for the vertical median filter.

Step 456—Vertical line detection: Within a sliding rectangular window, the brightest pixel in each row is displaced into the column containing the majority of brightest pixels, and each other pixel in the row is set to zero. The effect of this filter is to enhance vertical lines, and to convert the grayscale image into a binary image of candidate vertical edges. The resolution and contrast of real vertical edges improves, although some random noise begins to look like vertical lines. Various rectangle sizes can be used in this step. For the example in FIG. 3, a rectangular window of height 30 pixels (the average height of a chip, hence the minimum length of real vertical edges) and of width 5 pixels (the average error of the vertical edge detector) was used. In a higher resolution or more tightly focused image, larger rectangles would be used. In a lower resolution or more distant image, smaller rectangles would be used. The size of the optimal line detection rectangle is fixed by the properties of the imaging hardware, and remains substantially constant for a particular embodiment of the invention.

Step 458—Vertical path discovery: Locate continuous vertical paths that are at least as long as the height of a chip (30 pixels, for the example in FIG. 3), and which may connect with other vertical paths within a horizontal distance determined by the expected maximum misalignment of chips in a stack (30 pixels, for the example in FIG. 3). Pixels which are part of such vertical paths are set to binary 1's, and pixels which are not are set to binary 0's.

Step 460—Horizontal edge detection: Each pixel in the original color image is replaced by a grayscale value computed from the average difference between the color values of a number of pixels above it and the color values of a number of pixels below it. For the example in FIG. 3, three pixels were used for horizontal edge detection.

Step 462—Horizontal median filters: Each pixel in the horizontal edge image after step (e) is replaced by the median of a horizontal row of N pixels, of which it is the center. The effect of this filter is to eliminate horizontal lines shorter than N/2, and to fill in gaps in longer horizontal lines. Appropriate values of N depend on the image resolution and on the size of horizontal boundaries to be detected—values of N from 3 to ½ of the chip width may be optimal. For the example in FIG. 3, 11 pixels were used for the horizontal median filter. This is the horizontal distance in the image over which there is no appreciable curvature of the chip edge.

Step 464—Horizontal line detection: Within a sliding rectangular window, the brightest pixel in each column is displaced into the row containing the majority of brightest pixels, and each other pixel in the column is set to zero. The effect of this filter is to enhance horizontal lines, and to convert the grayscale image into a binary image of candidate horizontal edges. The resolution and contrast of real horizontal edges improves, although some random noise begins to look like horizontal lines. Various rectangle sizes can be used in this step. For the example in FIG. 3, a rectangular window of height 5 pixels (the average error of the horizontal edge detector) and of width 340 pixels (the average width of a chip, hence the maximum length of real horizontal edges) was used. In a higher resolution or more tightly focused image, larger rectangles would be used. In a lower resolution or more distant image, smaller rectangles would be used. The size of the optimal line detection rectangle is fixed by the properties of the imaging hardware, and remains substantially constant for a particular embodiment of the invention.

Step 466—Horizontal path discovery: Locate continuous horizontal paths that are at least as long as the uncurved width of a chip (100 pixels, for the example in FIG. 3), and which may connect with other horizontal paths within a vertical distance determined by the expected maximum curvature of chip edges (1 pixel/100 pixels, for the example in FIG. 3). Pixels which are part of such horizontal paths are set to binary 1's, and pixels which are not are set to binary 0's.

Step 468—Boundary detection: The vertical boundary lines from step 468 are combined with the horizontal boundary lines from step 466, to search for candidates for fully bounded rectangular regions. Vertical lines extending beyond their intersections with horizontal lines are truncated. Likewise, horizontal lines are truncated at corners. The result of processing the image of FIG. 3, after boundary detection, is shown in FIG. 5.

Step 470—Flood fill regions: Enclosed regions after step 468 are flood filled. Filled regions with dimensions less than ½ of a nominal chip dimension (height 30 pixels, width 320 pixels, in the image of FIG. 1) are eliminated.

Step 472—Clip stack candidates: The bounding rectangle of each filled region from step 470 is a candidate for a stack of chips. The five candidate stacks identified in the FIG. 3 image are shown in FIG. 6. The two smaller regions are rejected as too small to be chip stacks, leaving three candidate stacks located, as shown in FIG. 7.

After stack location, the original image has been replaced by zero or more smaller images, each including only a subset of the original image that depicts a stack of chips, as shown in FIG. 7. If no stacks of chips are found, the method terminates and reports that it was unable to detect any chips in the betting area. This might happen because of a failure of the imaging hardware, a failure of the stack location method, or improper placement of a stack by the player, or it might be a normal expected condition at the current stage of the game. In any case, the ABRS system in which the ABRS-IPPR software program is embedded takes appropriate action to either alarm the dealer or casino personnel, or perhaps to identify that player position as uninhabited.

It will be readily understood by those skilled in the art that there are many other combinations of digital filters, method steps, and pattern-matching steps known to the art which could be employed to achieve essentially the same stack location result as the exemplary sequence of digital filters and method steps described above and depicted in FIG. 4A-B. It is understood that the present invention includes any and all such combinations of filters known to the art, and is not limited to the exemplary sequence of filters described above.

If the stack location method detects one or more stacks of chips, subsequent steps in the method are typically performed separately and sequentially on each such stack. The values of all the chips in each stack are combined together for the final bet tabulation of step 410.

3. Locate Individual Chips in a Stack of Chips

The input to this step of the exemplary method is an image of a single stack of candidate chips, or one of the plurality of stacks, as shown in FIG. 7. The next task is to divide the image into separate images, each one an image of the edge of a single chip in the stack.

Although many algorithmic methods and combinations of algorithmic methods could be employed to locate stacks of chips in an image, the preferred embodiment comprises six sequential steps as shown in the FIG. 8 flowchart 800.

Step 802—As for the stack location method, a horizontal line search for parallel horizontal boundaries spaced close to the known chip thickness, combined with a regional continuity search, can find many chip boundaries.

Step 804—As for the stack location method, a horizontal median filter accentuates horizontal lines in the image. The initial separated image of an exemplary stack of chips is shown in FIG. 9A, and the result of the horizontal edge filter and edge enhancement is shown in FIG. 9B.

Step 806—The average edge intensity in each row is determined, and a peak-finding method is used to locate the average vertical position of horizontal edges. Frequently, there will be no clear boundary between two adjacent chips of the same denomination. In this case, the method uses its knowledge of the chip thickness (or aspect ratio) to deduce that two chips are stacked rather than just one. In this case, an edge between two peaks of approximately double (or triple, etc) the expected chip height is inferred. It is unnecessary that the software identify a precise boundary between adjacent identical chips—it is sufficient to know how many there are. In the example shown in FIG. 9A, the expected chip height is 30 pixels, so detected “peaks” less than 25 pixels apart, caused by shadows in the image (faint diagonal lines crossing chips in FIG. 9B), are ignored.

Step 808—Starting at the average vertical position of each inferred chip boundary, a dynamic programming algorithm, known as a Viterbi algorithm, is employed to determine the most likely path of the boundary between adjacent chips. The chip boundaries determined by this method for the chip stack image of FIG. 9B are shown in FIG. 9C.

Step 810—The chip boundaries detected in step 808 are curved, especially for chips near the top and bottom of the stack, because of parallax of images acquired at close range. In order to avoid including portions of neighboring chips in the individual chip images, the chips discovered in step 808 are clipped vertically, at positions such that about 80% of the horizontal boundary of the chip is excluded from the chip image.

Step 812—The boundaries of individual chips in the image, isolated in steps 802-810, are recorded for input to the chip identification method.

It is usually unnecessary for the chip location method to use any knowledge about the precise colors and/or patterns of the valid chip denominations. However, if the ambient lighting is especially uneven, or the chip patterns of different denominations of chips are insufficiently distinctive, the use of expected chip colors and patterns may increase the robustness of chip location.

The chip location method may be unable to subdivide the stack image into individual chip images, either because the stack was misidentified (e.g. it is really a similar-looking foreign object), or the stack is poorly aligned or contains foreign objects, or the ambient lighting or shadow obscures the chip boundaries. In some such cases, the “stack” can be rejected as a foreign object, and processing can proceed on other detected stacks. In the exceptional case in which a stack can neither be properly segmented nor rejected, the ABRS-IPPR notifies the parent ABRS system that it is unable to evaluate the bet, so the dealer can make a manual evaluation of the bet.

The final result of applying the method of flowchart 800 to the stack of FIG. 9A, is the set of individual chip images shown in FIG. 10.

It will be readily understood by those skilled in the art that there are many other combinations of digital filters, algorithmic steps, and pattern-matching steps known to the art which could be employed to achieve essentially the same chip location result as the exemplary sequence of digital filters and algorithmic steps described above and depicted in FIG. 8. It is understood that the present invention includes any and all such combinations of filters known to the art, and is not limited to the exemplary sequence of filters and algorithmic steps described above.

If the chip location method detects one or more chips in the stack, subsequent steps in the method can be performed separately and sequentially on each such chip. The values of all the chips in the stack are combined together, and with the counts and values obtained from other stacks, for the final bet report.

4. Tabulate Vertical Bands of a Single Chip

The input to this step of the method is an image of the edge of a single gaming chip, as shown in FIG. 12A. The next task is to characterize the colors, widths, and patterns of the several distinct regions of the chip edge. In general, each denomination of chip that can be legitimately wagered on a particular table has a unique combination of at least two distinct colors, arranged periodically around the periphery of the chip such that, regardless of the orientation of the chip, at least one complete instance of the alternating pattern of colors is visible in the image (which captures ½ of the periphery of the chip). In addition to alternating bands of different colors, the edge of the chip may display geometric patterns such as dots, bars, or other figures of one color, surrounded by regions of another color.

The band tabulation method characterizes each band of the chip visible in the image. A band consists of a roughly rectangular region of the image, spanning the full thickness of the chip, and demarcated on the left and/or right by regions of contrasting color. In some chip patterns, the boundary between bands is not vertical but V-shaped, slanted, or irregular. The method treats such band boundaries as if they were vertical. A flowchart of a preferred embodiment of the band tabulation method is shown in FIG. 11.

A preferred embodiment of the vertical band tabulation step comprises the following steps, as depicted in FIG. 11 flowchart 1100:

Step 1102—Vertical edge detection: As for the stack location method, a vertical line search finds candidate band boundaries.

Step 1102—Peak-picking: The average edge intensity in each column is determined, and a peak-finding method is used to locate the average horizontal position of vertical edges. The output of such an method is shown in FIG. 12B. Some filtering is applied to eliminate very small regions between “peaks.” The surviving peaks delimit regions of distinct color, which are candidates for the distinctive color bands defining the denomination of the chip.

Step 1102—Find the average color of each vertical band: For each region, compute the mean color, in each RGB channel, of the pixels within the region, excluding boundary pixels.

Step 1102—In some gaming chip designs, the color bands include dots, crosses, or other symbols of contrasting color. The ABRS-IPPR system can learn which chips have such enclosed symbols during training, as described below. However, unexpected enclosed symbols may be encountered, if unauthorized chips are present. The preferred method can detect such enclosed symbols using a boundary detection and flood fill method similar to that described in step 470 above. Enclosed symbols are counted and recorded for use in subsequent chip identification steps. Any such enclosed symbols are treated as separate regions in subsequent steps—that is, the average color and relative image area of each enclosed region is calculated. However, pixels in enclosed regions are excluded from the average color determination of step 1102.

Step 1102—Merge adjacent regions of similar color: If two adjacent regions have similar colors, they are presumed to differ by the intervention of a shadow, and should be merged into a single region. Bands surviving this step are shown in FIG. 12C.

Step 1102—Normalize brightness: Because the incident light falling on each region is variable, the relative intensity in the three colors is more distinctive than the absolute intensity of any color. This is accounted for by normalizing the total luminance of each pixel. This is done by normalizing the red (R) and green (G) values of each band such that the total luminance is 1.0. This normalization converts both black and white bands to grey; it is equivalent to removing the intensity dimension of an intensity/hue/saturation (ihs) color representation.

Step 1102—Find edges of the chip: The input images to flowchart 1100 are bounded by areas of background, which have nothing to do with the chip identity. Regions at the extreme left and right of the chip image are likely to consist of such irrelevant background. Color regions at the extreme left and right of the chip image are eliminated, unless they are large enough to extend beyond the possible boundary region.

Step 1102—Compute region size: Of the surviving regions, the true size (measured as degrees of arc) of each region is computed, based on the viewing angle. Thus, regions near the edge of the chip span fewer pixels in the image than do similar-sized regions near the center of the image. A simple approximation assuming an infinite viewing distance is used to normalize the region size.

Step 1102—The first and last regions identified by the above steps, unless they exceed a parameterized value, are excluded from the following analysis as unreliable data. The final normalized bands in the example chip of FIG. 12A are shown in FIG. 12D.

It will be readily understood by those skilled in the art that there are many other combinations of digital and heuristic filters, and algorithmic steps known to the art, which could be employed to achieve essentially the same band identification result as the exemplary sequence of digital filters and algorithmic steps described above and depicted in FIG. 11. It is understood that the present invention includes any and all such combinations of filters and algorithmic steps known to the art, and is not limited to the exemplary sequence of filters and algorithmic steps described above.

The output of flowchart 1100 is a list of all bands in the chip image, each characterized by (a) its true radial width, (b) its average normalized color, (c) the number, color, and relative size of any enclosed regions.

5. Tabulate Unique Bands

The input to this step of the method is a list of all bands detected in the chip image, their normalized colors, and the size, number, and color of enclosed symbols, if any. In this aspect of the method, the list of observed bands is reduced to a minimal ordered list of bands. The method searches for repeating patterns of sequences of bands of approximately the same color, width, and included regions. This minimal ordered list of bands constitutes the raw signature pattern of the chip.

If there are multiple examples of bands which are similar in color and size, which are consolidated into a single band in the above step, then these bands are retained in memory as potentially distinct bands. These potential additional bands may be used in subsequent steps of the method to allow matches with signatures containing additional bands.

If there are included symbols within the regions, then the number, colors, and relative size of such symbols is included in the signature pattern.

6. Standardize Colors

The input to this step of the method is an ordered list of all unique bands detected in the chip image. Before attempting to match this signature pattern to one of the expected authorized chip signatures, the colors are transformed into a maximally distinctive color space. The parameters of the color transformation are determined when the signature patterns of all authorized denominations of chips used at the gaming table are specified during training, as described below. After color transformation, the ordered list of unique band widths, transformed colors, and the number, color, and size of included regions, constitutes the standard signature pattern of the chip.

The standard signature pattern of the chip imaged in FIG. 10a is shown as a color image and in the table in FIG. 13.

It should be noted that the chip depicted in FIG. 12 in fact has three distinct color bands. The first band detected (“Orange2”, [0.74, 0.1, 0.14]) is actually the same as the fourth band ((“Orange2”, [0.61, 0.18, 0.10]). The standard signature patterns of these bands, actually identical, are measured as different because the first is lying in the shadow of other chips in the stack. Such variations of measured signatures under different conditions of light and shadow is to be expected. The exemplary method can nonetheless correctly identify gaming chips despite such variations, provided that the variation in lighting is not too severe, and the color patterns of authorized chips are sufficiently distinctive.

7. Match Chip Signature to Best Authorized Chip Signature

Once the standard signature pattern of each gaming chip has been determined, it is then classified as either one of the authorized chip patterns, or as an unauthorized pattern. Any of a several classification methods may be employed to achieve this classification, including linear classifiers, Bayesian classifiers, hierarchical classifiers, neural network classifiers, and others. While any of these or other classification methods could be used, a preferred method is a radial basis function neural network classifier. The advantage of a radial basis function neural network classifier over many other classifiers is that it is able to determine when a chip is not a member of the authorized set, without having to be explicitly trained to recognize unauthorized chips.

The standard signature pattern of the chip is supplied as input to a neural network, which computes the probability that the chip matches an authorized chip on which the network has been trained, as explained below. An example of a neural network suitable for such a determination is provided in FIG. 14. It should be readily apparent to one skilled in the art that there are many other neural network architectures and other classification methods that could be used to obtain substantially equivalent results to those obtained from the neural network of FIG. 14.

The neural network depicted in FIG. 14 comprises three layers: an input layer, a “hidden” layer, and an output layer. Each node in each layer is connected with each other node in the layer below it. In general, the links between a layer and the next layer are associated with a “weight,” which is a variable parameter multiplying the value of the node in the first layer. Each node computes a “transfer function,” a single valued function of the weighted inputs to the node, plus an additional added “bias” associated with the node. The value of the transfer function, for specified inputs, is then propagated on to the next layer of the network.

A “radial basis function” neural network is one in which the nodes of the second, “hidden”, layer, are peaked functions (for example, Gaussian functions) centered at a position in the input domain specific to that node (in this example, in the three-dimensional space of red, green, and chip width). With each node is associated a center and a width, and the value of the node is the sum of the values of the peaked transfer function when applied to the inputs. A preferred embodiment of the method uses radial basis functions which are three-dimensional Gaussian functions of the red and green values, and the radian dimensions, of each of the color bands represented in the input layer. With each node of the RBF layer there are associated six parameters: the red, green, and width coordinates of the center of the Gaussian function, and the red, green, and width standard deviations of the function. In this example, the standard deviations of the RBF functions in the input dimensions were constrained to be equal, so that there are four independent trainable parameters associated with each node in the hidden layer. The red and green dimensions of the radial basis functions in the network of FIG. 14, after training on the input denominations represented in the stack of gaming chips shown in FIG. 3, are shown in FIG. 15. Each peak in the figure is centered at the (red, green) standardized coordinates of one of the colors of one of the bands of the authorized chips. The value of the corresponding node of the network, when presented as input with the standardized chip signature of a chip, is the height of the RBF at the red and green coordinates observed for any color band of that chip.

The values of the output nodes (the third layer of the network depicted in FIG. 14) are computed as a weighted sum of the values of the nodes of the RBF layer, transformed by the transfer function of the third layer. In the preferred embodiment, the weights of the links between layers 2 and 3 are trained, and the layer 3 transfer function is a sigmoid transfer function (e.g. tanho), which then reports membership probabilities. Various other transfer functions and weight vectors would also be suitable for use in this method.

The output of the neural network depicted in FIG. 14, or of an alternative neural network or equivalent classifier, is the probability that the gaming chip is of a particular denomination of authorized gaming chips included in the set of chips on which the network was trained, as described below.

FIG. 16 shows the output produced by the neural network of FIG. 14, when presented as input with the signature patterns of seven of the chips from the example stack of FIG. 9. FIG. 16 includes three chips of denomination $5, two chips of denomination $1000, and two other chips that weren't included in the set of authorized chips used to train the network.

If the signature pattern of the chip matches exactly one of the authorized signatures within a specified tolerance, the software program records a chip of the corresponding denomination, and adds it to the amount of the bet.

If the signature pattern of the chip matches more than one of the authorized signatures within a specified tolerance, the ABRS-IPPR notifies the parent ABRS system that it is unable to determine the value of a chip in the stack. This can occur if irregularity of lighting or shadow causes the system to incorrectly estimate the color of a band in the chip signature. The ABRS can then signal the dealer to manually record the value of the bet. If this problem recurs, it may be necessary to reduce the number or increase the distinctness of the authorized chips at the table.

If the signature pattern of the chip fails to match any of the authorized signatures within a specified tolerance, the ABRS-IPPR notifies the parent ABRS system that an unauthorized chip is present in the stack. The ABRS can then signal the dealer to manually verify the irregularity, and take appropriate action. Whereas the game deal may choose to manually enter the amount of the wager into the system by means of a keypad as he/she would do in the event the players wager was cash or a call bet.

It will be readily understood by those skilled in the art that there are other arrangements and architectures of neural networks, and similar pattern-matching methods, which can similarly classify signature patterns into trained categories. It is understood that the present invention includes any and all such combinations of classification methods known to the art, and is not limited to the exemplary neural network architectures described above.

8. Report the Total Amount of the Wager

The total value of the player's bet is determined by summing the value of each chip and denomination thereof in each wagered stack. The ABRS-IPPR reports the total value of the chips in the player's betting area to the parent ABRS system.

C. Training to Determining the Authorized Chip Patterns

The invention also teaches a method of determining the signature patterns of chips that are authorized for a particular table in a casino, and determining the matching tolerances for identifying players' chips as one of the authorized denominations. The ABRS can distinguish among at least 12 different values (e.g. $1, $5, $25, $100, $500, $1,000, $5,000, $10,000, $25,000, $50,000 & $100,000) of authorized chip signatures. The reliability of the ABRS-IPPR system is dependent on the number and distinctness of the chip signatures. When the number of authorized signatures is about one dozen, and the colors of bands in the signatures are sufficiently distinct over the range of illumination in the casino, the method of the invention is quite reliable. As the number of authorized chip signatures increases, the method becomes less robust.

Prior to use of the system in any casino, the ABRS-IPPR is trained to recognize the chip signatures authorized for use at each gaming table. Such training is effected by capturing multiple images of stacks of chips, including all authorized denominations, over a range of illumination conditions representative of the conditions expected when the system is used. For each training stack imaged, the correct denomination of each chip in the stack is supplied to the training software program.

During training, the stack of chips of known denominations is imaged in the same orientation and configuration as would be used for determination of the values of a stack of actual gaming chips in the exercise of the methods taught in this invention. The true value of the chips in the stack, and, optionally, the true colors of the bands of said chips, are supplied to the training method. The training method uses methods well known to the art, such as error back-propagation, to iteratively adjust the parameters of the neural network so as to optimize the rate of correct classification of the chips in the training set.

During training, the relative size and the colors of enclosed symbols imprinted within color bands of the chips are recorded and included in the training set.

As a part of the training, the optimal color transformation of color bands of the authorized chips, to be used in the flowchart 800, is determined. The optimal transformation is such as to maximize the separation, in normalized color coordinates, of the chips in the authorized set. Techniques for choosing transformations which maximize the discrimination of exemplars in the training set are known to the art.

The number of training examples required to fully train the recognition system depends on the number and distinctness of the authorized chip denominations, and the variability of the illumination. In most cases, the number of training instances required will be between 100 and 10000. A sufficient number of training instances can be obtained quite readily—for example, if two stacks, each containing 10 chips of a chip denomination to be trained, were stacked in the betting area of each of five player positions at a gaming table, then a single image captured from each player position would provide 100 training instances.

During training, the ABRS-IPPR learns the average signature of each authorized denomination of chip, the optimal color transformation to be employed in flowchart 800 of the recognition method, and the optimal matching tolerance to be used when matching the candidate chip signature in the recognition method.

Increasing the number of training examples improves the ability of the system to correctly identify unauthorized chips. The training is typically non-linear, where the initial training has great impact and training in high volumes has an increasingly smaller impact, but which may nonetheless be worthwhile depending on the granularity of desired detection.

If any new chip pattern is authorized for play, or if the pattern of any authorized chip changes, the system should be retrained. It may be desirable to periodically retrain the system, even if there are few or no changes in authorized chip denominations or patterns, to protect against changes in the colors of different manufacturing lots of nominally identical chips, and against changes in illumination at the casino gaming tables.

D. Additional Applications

The invention is also applicable to other areas of a casino where it would be useful to identify chips. For example, the invention can be implemented in a cashier's booth to ensure that the correct amount of money is given to patrons in exchange for their chips. A screen can be provided visible to the cashier to inform the cashier of the correct total vale of the chips to be cashed in by the patron.

The invention can also be used in more than just card games, for example, craps and other such games where the player has an area to place a bet, the invention can identify the bet and assist the dealer in assessing the proper amount to pay out in the event of a player's win. In fact, since such games can include odds-related payouts requiring a calculation, the computer may be able to perform the calculation faster and more accurately than a dealer.

E. Player's True Worth Computation

The invention can be used to automatically determine a player's true worth to the casino. This computation is performed by the casino computer 150 using information regarding a player's wins and losses. Since the invention includes a player tracking card reader, the invention communicates the identity of the player to the casino computer along with the player's wins and losses. The casino then computes the value of the player to the casino and can provide the player with complementary goods and services based on the player's true worth.

As mentioned above, the invention can be used in combination with other peripherals such as a card shoe of the type described in U.S. Pat. Nos. 5,362,053; 5,374,061; 5,722,893; 6,039,650; 6,299,536 and 6,582,301 all incorporated herein by reference in their entirety. A card shoe of this type provides one type of means for signaling the beginning of a game. In one aspect, the dealer presses a button on the card shoe 106, chip tray 120 or other location to signal the start of the game to the computer. In another aspect, the card shoe 106 can automatically signal the computer 110 that a new game is beginning and can automatically signal the computer when the game is completed. This can be performed because the shoe tracks the cards dealt to the players and the dealer, and the shoe knows when a player wins and loses each hand and when the game is over.

The invention can be employed in combination with any intelligent shoe to track the player's winnings and losses for the casino. This can assist the casino in determining a player's true worth to the casino. Moreover, since the invention includes a player tracking card reader, the invention can transmit data to the casino computer regarding the player's bet and wins and losses. This way, the casino can track the player's wins and losses at many games in the casino, and can accurately determine the player's true worth.

F. Conclusion

The invention provides numerous aspects and advantages to a card game with automatic bet recognition. Advantages of the invention include the ability to identify chips wagered by a player to automatically determine the player's bet.

Having disclosed exemplary embodiments and the best mode, it will be understood by those skilled in the art that changes in form and detail may be made therein without departing from the spirit and scope of the invention.

Claims

1. A bet recognition system for use with a card game comprising:

a table for playing a card game including a bet location for each player;

an image capture device positioned in proximity to the bet location and configured to capture an image of a player's bet; and

an image processor coupled to the image capture device and configured to process the image by locating at least one chip and creating a signal representing the chip, comparing the signal to a plurality of stored signatures, when a match occurs, generating a signal representing the bet.

2. The bet recognition system of claim 1, wherein:

the player's bet includes a plurality of chips having at least two different denominations; and

the image processor is configured to identify edges of the chips, segment the image into a plurality of individual candidate chips, generate a signature for each of the candidate chips, identify each of the candidate chips by comparing the signature of each candidate chip to a plurality of stored signatures representing valid chips and associated denominations, and add denominations associated with each of the valid chips to determine the wager.

3. The bet recognition system of claim 1, wherein:

the image processor is configured to generate an error signal when unable to match a candidate signature with at least one of the stored signatures.

4. The bet recognition system of claim 2, wherein:

the image processor is configured to generate an error signal when unable to match a candidate signature with at least one of the stored signatures.

5. The bet recognition system of claim 1, further comprising:

means for determining the start of the game and the end of the game.

6. The bet recognition system of claim 2, further comprising:

means for determining the start of the game and the end of the game.

7. The bet recognition system of claim 3, further comprising:

means for determining the start of the game and the end of the game.

8. The bet recognition system of claim 4, further comprising:

means for determining the start of the game and the end of the game.

9. The bet recognition system of claim 1, further comprising:

a chip tray image capture device positioned in proximity to a chip tray and configured to capture an image of the contents of the chip tray; and

an image processor coupled to the chip tray image capture device and configured to process the image by locating at least one chip and creating a signal representing the chip, comparing the signal to a plurality of stored signatures, when a match occurs, generating a signal representing the contents of the chip tray.

10. A method of processing an image representing a plurality of chips having a plurality of different denominations, comprising the steps of:

capturing an image representing the plurality of chips;

identifying edges of the plurality of chips;

segmenting the image into a plurality of individual candidate chips;

generating a signature for each of the candidate chips;

identifying each of the candidate chips by comparing the signature of each candidate chip to a plurality of stored signatures representing valid chips and associated denominations; and

adding denominations associated with each of the valid chips to determine the wager.

11. The method of claim 10, further comprising the step of:

generating an error signal when unable to match a candidate signature with at least one of the stored signatures.

12. The method of claim 10, further comprising the step of:

determining the start of the game and the end of the game.

13. The method of claim 11, further comprising the step of:

determining the start of the game and the end of the game.

14. The method of claim 10, further comprising the steps of:

capturing an image representing chips in a chip tray;

identifying edges of the plurality of chips;

segmenting the image into a plurality of individual candidate chips;

generating a signature for each of the candidate chips;

identifying each of the candidate chips by comparing the signature of each candidate chip to a plurality of stored signatures representing valid chips and associated denominations; and

adding denominations associated with each of the valid chips to determine the contents of the chip tray.

15. A method of training an image processor to process an image representing a plurality of chips having a plurality of different denominations, comprising the steps of:

capturing a plurality of images representing chips having different known denominations;

generating a signature for each of the chips having different denominations; and

storing the signatures.

16. The method of claim 15, wherein:

the capturing step includes capturing a plurality of images for each of the chips having different denominations; and

the generating step includes generating at least one signature for each of the images of the chips having different denominations, thereby generating a plurality of signatures for each of the chips having different denominations;

17. The method of claim 15, wherein:

the generating step includes generating a plurality of signatures for each of the chips having different denominations.

18. The method of claim 16, wherein:

the generating step includes generating a plurality of signatures for each of the chips having different denominations.