Abstract: The invention is a method and apparatus for improving image-processing applications. Embodiments of the invention provide methods for preserving computation results and offer intermediary computation steps to allow the processing of images at any location to take advantage of previously processed image areas. The preferred embodiment offers a method for building a median filter that significantly improves the processing speed over basic techniques. By building a histogram hierarchy, image data statistics are added and subtracted using a multiplicity of histograms from the histogram hierarchy, where each histogram describes an image area. Furthermore, a histogram hierarchy is built using multiple layers, each layer defining a level of statistical resolution.

Abstract: The present invention is directed to method of detecting a scene change in video. The method includes a difference being calculated between consecutive frames of the video producing a plurality of difference values. A histogram is taken of the difference values. A scene change is detected if a majority of the difference values in the histogram exceed a predetermined value.

Abstract: A system and method to facilitate pattern recognition or matching between patterns are disclosed that is substantially invariant to small transformations. A substantially smooth deformation field is applied to a derivative of a first pattern and a resulting deformation component is added to the first pattern to derive a first deformed pattern. An indication of similarity between the first pattern and a second pattern may be determined by minimizing the distance between the first deformed pattern and the second pattern with respect to deformation coefficients associated with each deformed pattern. The foregoing minimization provides a system (e.g., linear) that may be solved with standard methods.

Abstract: An apparatus for reproducing an information signal stored on a first storage medium (4). The apparatus comprises a reading unit (2) for reading the information signal from the first storage medium, an output unit (6) for supplying the information signal to a display unit, an user controllable input unit (8) for receiving commands to enable an user to access the information signal. The user controllable input unit is adapted to receive a first command at an instant. The apparatus further comprises a unit (10) for controlling the reading unit to start reading the information signal from said storage medium at a second position in the information signal in response to said first command, the information signal at said second position having features showing a similarity with features of the information signal at a first position read at said instant of receiving said first command, or a features of an information signal read prior to said instant.

Abstract: An apparatus and method for inspecting parts. The apparatus includes an x-ray source for illuminating a part from a plurality of locations with respect to the part and an imaging detector for forming a plurality of measured x-ray images of the part, one such measured x-ray image corresponding to each of the illumination locations. A controller compares each of the measured x-ray images with a corresponding calibration image. The controller provides a defective part indication if one of the measured x-ray images differs from the corresponding calibration image by more than a threshold value in part of the measured x-ray image. The controller localizes defects on the part by comparing two or more of the measured x-ray images with two or more corresponding calibration images. The calibration images can be constructed from measured images of defect free parts.

Abstract: A character reading technique recognizes character strings in grayscale images where characters within such strings have poor contrast, are variable in position or rotation with respect to other characters in the string, or where portions of characters in the string are partially obscured. The method improves classification accuracy by improving the robustness of the underlying correlation operation or the character design. Characters are divided into regions before performing correlations. Based upon the relative individual region results, region results are combined into a whole character result. Using the characters that are read, a running checksum is computed and, based upon the checksum result, characters are replaced to produce a valid result.

Abstract: A method for representing an image in terms of the shape properties of its identified segments of interest involves analyzing each segment of interest to automatically identify one or more of its perceptually significant components and then representing each perceptually significant component in terms of its shape properties, e.g., by means of Fourier descriptors. The image segment is then characterized as a composition of the shape properties of its perceptually significant components. By repeating the foregoing steps, the image is represented as a composition of its characterized segments. This technique can then be used to archive and retrieve images based on the shape properties of identified segments.

Abstract: A system and method to facilitate pattern recognition or matching between patterns are disclosed that is substantially invariant to small transformations. A substantially smooth deformation field is applied to a derivative of a first pattern and a resulting deformation component is added to the first pattern to derive a first deformed pattern. An indication of similarity between the first pattern and a second pattern may be determined by minimizing the distance between the first deformed pattern and the second pattern with respect to deformation coefficients associated with each deformed pattern. The foregoing minimization provides a system (e.g., linear) that may be solved with standard methods.

Abstract: A method for resolution conversion for re-sampling anti-aliased images is disclosed which decreases bandwidth costs associated with anti-aliased line art and other costs associated with interpolating these images to a desired resolution. The present method first involving the receipt of an image which is comprised of bitmap data including at least a plurality of gray-scale pixel tiles that define the image. Then receiving the image data at a first resolution and extracting pixel tile information of the received image at a second resolution. The method has the step of next using loose gray scale template matching on each of the pixel tile information with at least one of a plurality of templates so as to generate pixel-wise looseness interval values there between. Then, outputting a portion of the enhanced pixel tile information wherein the enhanced pixel tile information is formed by a matching of a template with pixel-wise looseness values.

Abstract: Techniques for classifying video frames using statistical models of transform coefficients are disclosed. After optionally being decimated in time and space, image frames are transformed using a discrete cosine transform or Hadamard transform. The methods disclosed model image composition and operate on grayscale images. The resulting transform matrices are reduced using truncation, principal component analysis, or linear discriminant analysis to produce feature vectors. Feature vectors of training images for image classes are used to compute image class statistical models. Once image class statistical models are derived, individual frames are classified by the maximum likelihood resulting from the image class statistical models. Thus, the probabilities that a feature vector derived from a frame would be produced from each of the image class statistical models are computed.

Abstract: A character reading technique recognizes character strings in grayscale images where characters within such strings have poor contrast, are variable in position or rotation with respect to other characters in the string, or where portions of characters in the string are partially obscured. The method improves classification accuracy by improving the robustness of the underlying correlation operation. Characters are divided into regions before performing correlations. Based upon the relative individual region results, region results are combined into a whole character result. Using the characters that are read, a running checksum is computed and, based upon the checksum result, characters are replaced to produce a valid result.

Abstract: A method discovers patterns in unknown multi-dimensional data. A time-series of the multi-dimensional data is generated and a point cross-distance matrix is constructed by self-correlating the time-series. All minimum cost paths in the point cross-distance matrix are located at multiple time resolutions. The minimum cost paths are then related to temporal sub-sequences in the multi-dimensional data to discover high-level patterns in the unknown multi-dimensional data.

Abstract: A method and apparatus for identifying scale invariant features in an image and a further method and apparatus for using such scale invariant features to locate an object in an image are disclosed. The method and apparatus for identifying scale invariant features may involve the use of a processor circuit for producing a plurality of component subregion descriptors for each subregion of a pixel region about pixel amplitude extrema in a plurality of difference images produced from the image. This may involve producing a plurality of difference images by blurring an initial image to produce a blurred image and by subtracting the blurred image from the initial image to produce the difference image. For each difference image, pixel amplitude extrema are located and a corresponding pixel region is defined about each pixel amplitude extremum. Each pixel region is divided into subregions and a plurality of component subregion descriptors are produced for each subregion.

Abstract: Configuration is made by a feature pattern extracting section 12 for extracting a feature obtainable from a person, a subspace calculating section 14 for calculating a subspace from the feature pattern, a recognition section 16 for effecting recognition using the subspace, a registered information holding section 18 for holding a plurality of the subspaces, and a registered information updating section 20 for updating the registered information using the subspaces. When identifying a person, the effect of aging change is reduced by suppressing time and year changes and keeping the recognition rate high.

Abstract: A method and a device for the identification of incorrectly orientated parts and/or parts departing from a predetermined master, the parts being moved by means of a conveyor means (13) past at least one camera (18) for registering the shapes of the parts. When there is a preset departure from stored masters values the respective part is identified as being faulty and/or removed. On the basis of image signals of the camera (18) corresponding to the shape of the respective part a plurality of criteria are derived such as overall length, overall height, area, axial position, surface waviness, position of center of gravity. Such individual compared with the corresponding individual criteria derived from at least one master part. This means that with a relatively small memory requirement the parts may be reliably identified and more particularly significant fine structures may be distinguished.

Abstract: An apparatus and method of template fitting for fitting a plurality of templates in correspondence with a gray-scale image, including receiving gray-scale image information of an observed image; defining a window of the gray-scale image information so as to enable a loose-gray-scale fit between the defined window and the plurality of templates; extracting a plurality of pixels of the image information and using the window to which to fit at least one of the plurality of templates. Further, fitting one of the plurality of templates to the defined window within a threshold looseness interval associated with the template wherein a looseness interval between the window of image data and the template is less than the threshold looseness interval.

Abstract: A pattern recognition matching processing method and apparatus uses expansion processing with dictionary data as a matching candidate (pattern image). Contraction processing may be performed for the pattern image, and a differential image between the expanded and contracted images can be found. A mask image is prepared based on the differential image, and matching processing of the image containing a character to be recognized with the pattern image is performed with respect to pixels in an area other than the prepared mask image.

Abstract: A digital image processing method determines the extent of blocking artifacts in a digital image by first forming a column difference image and averaging the values in the columns in the column difference image to produce a column difference array. The average of the values in the column difference array that are separated by one block width are computed to produce a block averaged column difference array. Then, the peak value in the block averaged column difference array is located, and the mean value of the block averaged column difference array (excluding the peak value) is calculated to produce a column base value, and the ratio between the peak value and the base value are computed to produce a column ratio. The foregoing steps are repeated in the row direction to produce a row ratio. Finally, the column and row ratios are employed as a measure of the extent of blocking artifacts in the digital image.

Abstract: A method for matching images is provided that includes the step of providing a template image and an input image. The images include pixels, each pixel having an intensity associated therewith. An energy function formed by weighting a sum of squared differences of data constraints corresponding to locations of both the template image and the input image is minimized to determine estimated geometric and spatially-varying illumination-parameters for the input image with respect to the template image. The estimated geometric and spatially-varying illumination parameters are outputted for further processing.

Abstract: Configuration is made by a feature pattern extracting section 12 for extracting a feature obtainable from a person, a subspace calculating section 14 for calculating a subspace from the feature pattern, a recognition section 16 for effecting recognition using the subspace, a registered information holding section 18 for holding a plurality of the subspaces, and a registered information updating section 20 for updating the registered information using the subspaces. When identifying a person, the effect of aging change is reduced by suppressing time and year changes and keeping the recognition rate high.

Abstract: A system and method for ascertaining blockage of one or more stencil apertures is provided. An analysis of pixels in an image of the stencil region that includes the aperture is made that includes determining the relative grayscale intensity of pixels acquired from the clean stencil and also the variance among nearby pixels. The grayscale and variance values for each pixel are combined to form a series of histogram groupings or bins, based upon levels. These values are used to compute parameters that represent how the apertures appear under the lighting conditions used to acquire the image, thereby allowing the informing of a decision function that estimates the probability that a given pixel represents unblocked aperture rather than solder paste.

Abstract: Identifying content and authorizing sharing thereof based at least in part on extracting feature data from content to be shared and comparing this extracted feature data against a database of extracted feature data for various reference content. When the content to be shared is matched to reference content in the database, sharing may be simply rejected, or rights associated with the reference content may be inspected to facilitate determining authorization.

Abstract: Structure-guided image estimation and measurement methods are described for computer vision applications. Results of the structure-guided estimation are symbolic representations of geometry entities such as lines, points, arcs and circles. The symbolic representation facilitates sub-pixel measurements by increasing the number of pixels used in the matching of image features to structural entities, improving the detection of structural entities within the image, weighting the contribution of each image sample to the measurement that is being made and optimizing that contribution. After the structure-guided estimation, geometric entities are represented by their symbolic representations. Structure-guided measurements can be conducted using the symbolic representation of the geometric entities. Measurements performed from the symbolic representation are not limited by image resolution or pixel quantization error and therefore can yield sub-pixel accuracy and repeatability.

Abstract: A character reading technique recognizes character strings in grayscale images where characters within such strings have poor contrast, are variable in position or rotation with respect to other characters in the string, or where portions of characters in the string are partially obscured. The method improves classification accuracy by improving the robustness of the underlying correlation operation. Characters are divided into regions before performing correlations. Based upon the relative individual region results, region results are combined into a whole character result. Using the characters that are read, a running checksum is computed and, based upon the checksum result, characters are replaced to produce a valid result.

Abstract: A system creates and uses a visual image index for visual image retrieval. The visual image index uses multi-level filtering to reduce the processing and storage needed to identify visual images in a table that have similar characteristics to those of a query image. Information about a visual image is stored in an image signature that contains data about basic visual primitives of the image (e.g., global color, local color, texture and structure). The signature is stored in a database table along with the image itself (or a link to the image). Visual image retrieval is performed by searching for signatures in the database that are similar to the signature of the query image. The multi-level filtering reduces the number of candidate images, before a full comparison is done on the filtered-set of candidate image signatures. The first sub-filter performs a range-based query on a table of image attributes culled from the image signature, each attribute in this table is indexed.

Abstract: A system and method is disclosed for determining a level of similarity among more than one image. Anticipated spatial characteristics of an image are used for automatically identifying segments within the image and for identifying weights to be added to the color characteristics associated with the identified segments. To determine similarity, comparisons are made between weighted color characteristics of corresponding segments of different images. The identified segments have attributes such as size, position and number which are based on the anticipated spatial characteristics of the image. The anticipated spatial characteristics of the image include, among other things, differences in image characteristics that are anticipated at relative positions within the image.

Abstract: In an image recognition method of recognizing a plurality of images from an input image, the plurality of images to be recognized including at least one of a specific image and images provided by rotating the specific image, the method includes the steps of: preparing at least one of mask areas on a window area for each image to be recognized, the window area being structured by cells which are arrayed in two-dimensionally, the cell being formed of at least one pixel, the mask area being structured by at least one of the cells and corresponding to an area representing a feature of the image to be recognized; first determining whether or not the image to be recognized exists based on the result of subjecting pixel density values of the input image in the mask area to operation for each image to be recognized; and second determining whether or not any of the plurality of images to be recognized exists in the window area based on the determination result of the first determining step.

Abstract: A method for localization refinement of inspection patterns comprises the steps of providing a template image comprising pixels in a pattern, each pixel having an intensity, providing an input image having a same pattern of pixels as the template image and calculating an energy function by weighting a sum of modified optical flow constraints at locations of the pixels of both the template image and the input image to determine a shift and rotation between the pattern of the template image and the input image.

Abstract: A sample image (142) is recognized by normalizing (404) the size of a sample image (142) to the size of a referent images (146); and determining (406) a set of candidate images (147) from a set of referent images (146), wherein each of the candidate images (147) is within an acceptable distance from a different binarization (145) of the sample image (142). A system (120) for image recognition includes a scanning device (126), a normalization unit (134), a distance calculation unit (136), a classification unit (138), a disambiguation unit (140), and a display device (128).

Abstract: A pattern data matching device comprises multiple target pattern data storing parts for storing target pattern data transferred from the matching controller, a file pattern data storing part for storing file pattern data transferred from the matching controller, and a data processing part for matching a necessary set of target pattern data among multiple sets of target pattern data stored in the multiple target pattern storing parts against file pattern data stored in the file pattern storing part.

Abstract: A tracking system and method for evaluating whether image information for a region cluster of a video frame of a scene represents an hypothesis of an object to be tracked, such as a person. At least one real-world feature of a region cluster corresponding to an object to be tracked is generated. For example, the at least feature is at least one possible location of a predetermined portion of an object represented by the region cluster is determined based on a viewing angle of the scene of the video camera. A distance from the video camera to the object corresponding to the region is determined in real-world coordinates for each possible location of the predetermined portion of the region cluster. Real-world size and location information for the region cluster is determined based on the distance.

Abstract: An apparatus for determining a similarity between model image and inputted image, which is robust even if the background of said inputted image is partly changed. The similarity is calculated from an edge direction of an inputted image, an edge direction of a model image and a weight related to the model image. The calculating steps used are (1) calculating an edge direction difference between an inputted model image and a model image, (2) calculating weights related to model image, (3) multiplying edge directions and weights, (4) calculating a sum of the multiplication, (5) calculating a sum of the weights, and (6) calculating a similarity by dividing the sum of the multiplication by the sum of the weights.

Abstract: The present invention has an object to determine whether image data is generated by data processing such as PDL, or optically captured scan-in image, for each area.To achieve the object, a determining table is provided for distribution patterns of pixels having substantially the same pixel value for image data area composed of plural pixels. Then a reference area designating unit designates a reference area by selecting from input image data the pixel of interest having a pixel value and the reference pixels surrounding thereof. A same pixel value distribution generating unit compares the pixel value of the pixel of interest with the pixel values of the reference pixels in the reference area to generate a distribution pattern of the pixel having substantially the same pixel value. Next, a determining unit compares the generated distribution pattern with the distribution patterns held in the determining table to determine the reference area to be the image data generated by data processing.

Abstract: A method of detecting whether a user is in close proximity to an end point. A base video frame of the end point is captured when the user is not in close proximity to the end point. A current video frame of the end point is captured and compared to the base video frame of the end point to determine whether the user is in close proximity to the end point.

Abstract: A statistical classifier for pattern recognition, such as a neural network, produces a plurality of output signals corresponding to the probabilities that a given input pattern belongs in respective classes. The classifier is trained in a manner such that low probabilities which pertain to classes of interest are not suppressed too greatly. This is achieved by modifying the amount by which error signals, corresponding to classes which are incorrectly identified, are employed in the training process, relative to error signals corresponding to the correct class. As a result, output probabilities for incorrect classes are not forced to a low value as much as probabilities for correct classes are raised.

Abstract: A method of encoding and decoding a transaction card with a block of image data digitally representing the features of a digital portrait of an authorized transaction card user is disclosed along with apparatus for performing the method. The encoding method partitions the digital portrait of the transaction card user into feature blocks. Each feature block is compared against a library of like feature blocks to determine the best match with a pair of library feature blocks. The library's feature blocks are derived from a plurality of portraits taken from the general population. Each library feature block is represented (addressed for access) by a code-vector in a codebook.

Abstract: The present invention relates to a new and useful Automated Pattern Recognition Device comprising a neural-network system, implemented on a general purpose computer, and capable of recognizing not only printed characters but also handwritten characters and other patterns in n-dimensions. The system incorporates novel feature extraction which generates an additional dimension from an n-dimensional input pattern, for example, a three-dimensional feature pattern from a two dimensional input pattern, resulting in shift-invariance, scale-invariance, and invariance to slight rotation.

Abstract: A statistical classifier that can be used for pattern recognition is trained to recognize negative, or improper patterns as well as proper patterns that are positively associated with desired output classes. A set of training samples includes both the negative and positive patterns, and target output values for the negative patterns are set so that no recognized class is indicated. The negative patterns are selected for training with less frequency than the positive patterns, and their effect on training is also modified, so that training is focused more heavily on positive patterns.

Abstract: Images in an image database are searched in response to queries which include the visual characteristics of the images such as colors, textures, shapes, and sizes, as well as by textual tags appended to the images. Queries are constructed in an image query construction area in response to values of representations of the visual characteristics and to locations of the representations in the image query construction area.

Abstract: Each pixel of a density reference area of a reference image is assigned a predetermined coefficient value in accordance with whether the pixel is located within a peripheral boundary of a to be detected shape, external the peripheral boundary of the to-be-detected shape, or on the peripheral boundary of the to-be-detected shape. An actual image of an object containing the to-be-detected shape is then obtained, and after setting a target pixel of the resultant image, the image density of each pixel is multiplied by the coefficient values of the density reference area of the reference image. The resultant products are transformed into a density correlation value for the target pixel. The location of the target pixel is then scanned to determine the location of a target pixel at which a maximum or a minimum density correlation value is obtained. The thus determined location is detected as the location of the to-be-detected shape within the object.

Abstract: Images in an image database are searched in response to queries which include the visual characteristics of the images such as colors, textures, shapes, and sizes, as well as by textual tags appended to the images. Queries are constructed in an image query construction area in response to values of representations of the visual characteristics and to locations of the representations in the image query construction area.

Abstract: A pattern recognition system is disclosed that uses a predetermined minimal number of comparison pixels in a pattern recognition process for input image recognition. For each pixel within a prescribed frame, a probability of the existence of a reference pattern is calculated by a probability calculation means. These probability values are used by a entropy calculation means to calculate pixel entropy values. Pixels with a high entropy value are then extracted and designated as comparison pixels for the pattern recognition operation. Comparison means compares these comparison pixels with corresponding pixels of an image pattern, and determines which of the reference patterns has a highest probability of being the image pattern.

Abstract: A pattern matching multiplies distances between an input pattern and standard patterns by distance-scale correcting weight coefficients respectively prepared for the standard patterns. This yields a distance corresponding to a category shape of a standard pattern and enhance an input pattern recognition ratio. A pattern recognition apparatus stores distance-scale correcting weight coefficients respectively prepared for all standard patterns in storage means. Distances between stored standard patterns and an input pattern generated by input pattern generating means are calculated. The calculated distance is multiplied by a weight coefficient stored in the storage means in accordance with the standard pattern used for calculating the distance from the input pattern. This structure reduces the number of templates when providing the standard patterns as multiple templates and, at the same time, enhances the input pattern recognition ratio.

Abstract: A method of inspecting a pixel for tolerable variation from a corresponding selected pixel in a master image comprises establishing a data word of which the bits each indicate whether or not all the master image pixels in the neighborhood of the selected pixel all have values (i.e. brightness values) within a respective range. A data word representing the value of the inspected pixel is compared simultaneously with the ranges of values associated with the bits of the data word, which provides gating for the comparisons.

Abstract: A method and apparatus for locating curvilinear objects of variable scale in digitized images using feathered fiducials. The leathered fiducials are constructed by: defining a range of tolerated object size scale variations; generating an enlarged submodel within the range of tolerated scale variations and assigning background pixel values exterior to the enlarged submodel; generating a reduced size submodel within the range of tolerated scale variations and assigning foreground pixel values to the reduced size submodel; disposing the reduced size submodel within and tangent to the enlarged submodel at the application point; uniformly changing grey scale pixel values in a graduated manner from foreground to background between the reduced size submodel and the enlarged one. The curvilinear objects are then located by conducting searches using the constructed fiducials.