Abstract: An image, having N dimensions, is partitioned into a set of bounded arbitrarily shaped regions. This partitioning is achieved through the use of arbitrarily shaped N-1 dimensional surfaces which pass through the boundaries of the image. The arbitrarily shaped regions are represented in the leaf nodes of a binary tree and the arbitrarily shaped surfaces, which are used to partition the image, are represented in the non-leaf nodes of the tree. In a preferred embodiment, arbitrarily oriented lines passing through the image boundaries are employed to partition a two-dimensional image into a set of convex regions. The convex regions are represented in the leaf nodes of a binary tree and the arbitrarily oriented lines, which are used to partition the image, are represented in the non-leaf nodes of the tree. Furthermore, the arbitrarily oriented lines are selected through the use of a Hough transform and the image is partitioned, along the selected lines, in a recursive manner as the binary tree is developed.

Abstract: The present invention features a method of characterizing texture of an image. The image is interrogated at least twice in accordance with at least two respective spatial threshold rules. A value for each of the interrogations is obtained. A mesotexture index is calculated by dividing the product of the values by the difference thereof. A plurality of resulting mesotexture indices can be used to generate a characteristic curve that uniquely identifies the image. Since the process is not iterative, a plurality of interrogating steps can be performed in parallel.

Abstract: A character recongintion device has a subdivider, a features calculator and a character code recognition device. Image data for a single character area is extracted from scanned character image data and input to the subdivider. This subdivider divides the image data for the single character area into subregions. The features calculator calculates quantified features in each subregion based on a degree of resemblance between a template and image data in the subregions. When the features of each subregion are calculated for all subregions constituting the single character area, a character code corresponding to the scanned character image data is recognized by the character code recognition device based on the quantified features of each of all subregions.

Abstract: A device for reading of (a) location(s) on an image on a transparent support and an apparatus for scanning and digitizing at least part of said image defined by said location(s) on the transparent support.

Abstract: An image scanner is provided that receives light reflected from a same position on a document by first, second, and third one-dimensional image sensors arranged in parallel. The second image sensor is a high-resolution image sensor comprising a plurality of photocells, which is disposed between and in a spaced-apart relationship with the first and third one-dimensional image sensors which are each low-resolution image sensors comprising photocells which are larger in size than the photocells of the second one-dimensional image sensor. Electrical image output signals from the first and third low-resolution image sensors are added, and any high-frequency components in the added signal are removed by a low-pass filter for producing signal data relating to the mean brightness of the region around a pixel of interest.

Abstract: An apparatus and method electronically registers or aligns ideal raster data to real raster data for subsequent processes such as inspection systems and image processing systems. The electronic image registration system receives ideal and real images in scrolling buffers wide enough to hold the width of the image and high enough to hold the difference in a skewed image. In other words, the real and ideal upper corners must be contained within the buffers for any possible kind of misalignment or distortion. Each buffer has a selector to randomly access frames of image in the vertical direction. Each buffer image frame is moved horizontally out of its lower swath and into its new position in the next upper swath. Horizontal movement of each buffer is independent, such that adjustable sequential access in the horizontal direction is available.

Abstract: Segmentation of characters in a character set (10), made by placing a dark mark against a light background (12), is accomplished by establishing a vertical pixel projection for each pixel column in the image. The vertical pixel projections are filtered with a decay parameter so those pixel columns which contain only background have the highest projection. Thereafter, a set of "cut-points" (points of image segmentation) is obtained so that each cut-point coincides with a pixel column whose vertical pixel projection is both a local maxima and exceeds a predetermined threshold. The number of such cut-points is counted and if the number is not significantly greater than a predetermined number, the image is segmented along the cut-points. Otherwise, the vertical projections of those pixel columns coincident with the cut-points are filtered with a decreasing threshold to reduce the number of potential cut-points.

Abstract: A method and apparatus for automatically verifying a signature of an individual is described. The apparatus digitizes an image of the signature and provides a grey scale representation of each pixel. Next, the apparatus determines parameters of this digitized image and compares them to corresponding reference parameter values which were determined from a valid signature. The results of the comparison indicate if the signature to be verified is valid. The parameters include a center of gravity of the pixels within each row and columns of the digitized image, a center of gravity line for the centers of gravity for the rows or columns and the center of gravity for all pixels of the image. Other parameters include the positions of maximum grey scale pixels within each row or column, sums of grey scale values per column or row and the shape of a bow within the signature image.

Abstract: An image processing area is divided into a plurality of small areas which are allowed to overlap one another. One or a plurality of pixels in a region to be segmented and in a region not to be segmented are specified. A representative point of the region to be segmented and a reference point of the region not to be segmented are determined from the color information of each of the pixels. A discriminating plane, a plane or a curved surface that intersects the line segment connecting the representative point and the reference point a right angles is used in processing. By the processing a particular region can be segmented out of the image. Further, a shape model is made by two-dimensionally projecting from a given direction the three-dimensional structure expressed by a line drawing consisting of nodes and curves connecting the nodes based on priori property relating the structure of the region to be segmented in the image.

Abstract: A novel lossy image compression algorithm--nonuniform sampling and interpolation, or NSI--for fast uncompression on a low power compute platform is disclosed. Decoder simplicity is traded off for a slight (2.5 dB) signal to noise ratio degradation over a non-adaptive discrete cosine transform (DCT) and increased encoder complexity. Encoding is three times slower than the DCT. High quality images are obtained at compression rates as high as 8:1 for 8 bit per pixel greyscale images. NSI can uncompress an image 48 times faster than a DCT. This speed advantage is traded for a small degradation in quality and higher compressing complexity. However, the characteristics of NSI make it ideal for use in an application (a) where images are more often uncompressed than compressed and (b) where a PC will be decompressing the image without expensive hardware assistance.

Abstract: Characteristics of images such as skew of lines of text or dominant typeface of characters are detected by producing distance data. Distance data is produced for each of a number of starting pixels within an image. Each starting pixel meets a criterion, such as an edge pixel or a pixel inside a connected component. Each starting pixel's distance data indicates the distance in each of a number of directions from the starting pixel to another pixel at which the image meets another criterion. For example, to detect skew of lines of text, the distance data can indicate distance from a starting pixel at an edge across white pixels to another edge. To detect dominant typeface, the distance data can indicate distance from a starting pixel at an edge or inside a connected component across black pixels to another edge.

Abstract: A self-organizing pattern classification neural network system includes means for receiving incoming pattern of signals that were processed by feature extractors that extract feature vectors from the incoming signal. These feature vectors correspond to information regarding certain features of the incoming signal. The extracted feature vectors then each pass to separate self-organizing neural network classifiers. The classifiers compare the feature vectors to templates corresponding to respective classes and output the results of their comparisons. The output from the classifier for each class enter a discriminator. The discriminator generates a classification response indicating the best class for the input signal. The classification response includes information indicative of whether the classification is possible and also includes the identified best class.

Abstract: A system and method for automatically processing machine printed forms which are filled out by hand are disclosed. The inventive system and method eliminate the use of a manual data entry step. Typically each form has one or more fields, each field having a machine printed field identifier such as the words "name" or "social security number" and the corresponding information which is filled in by hand. A scanner or other device forms a digital image of each form. An optical recognition engine searches the digital image of each form through use of an algorithm which can recognize machine printed but not handwritten characters to locate the fields in the digital image by locating the machine printed field identifiers. An offline handwritten character recognition algorithm is then utilized to recognize the handwritten characters in each individual field. Symbolic information corresponding to the recognized handwritten characters is then stored in a database in a memory.

Abstract: An image processing system including a plurality of sections disposed in an array. Under control of a control register, the function of a plurality of interconnected function generators is selected to process image data. By reason of the control register and the array structure, the number of different and unique image processing functions which can be produced is very large.

Abstract: Method for determining when a cigarette in a bundle is unsuitable which uses a statistical value based on gray scales.

Abstract: An image recognizer which recognizes an image signal representing a mark on a mark sheet which has been written in a predetermined format judges whether or not the image signal may be considered to be constituted in accordance with the predetermined format, and recognize only the image signal which is judged to be constituted in accordance with the predetermined format.

Abstract: A method in which the position of an object pattern in an image can be accurately detected is disclosed. The method comprises the steps of: (a) preparing a reference pattern which has a pattern and a background which are composed of pixels each having a value regarding the brightness or gradation; (b) dividing the image into a plurality of sections; (c) selecting arbitrarily an area having two or more of the sections; (d) obtaining the brightness data from each of the sections; (e) obtaining a sum of products of the value of each pixel and the brightness data of each section corresponding to the pixel; (f) moving the area to another position in the image; (g) repeating steps (d) to (f); and (h) deciding that the area from which the sum having an extreme value is obtained is the one containing the object pattern.

Abstract: A facsimile apparatus possesses a function in which the number of output copies of a received document is designated in the form of image data. In order to obtain this function, sheet information, on which character information is disposed in accordance with a predetermined rule, is received along with the document information. The character information of the received sheet information is recognized, and the number of output copies of the received document information is analyzed based upon the recognized character information. The received document information is outputted in accordance with the number of copies determined by analysis.

Abstract: A color image data processing apparatus processes hue information of picture elements read out sequentially. A chromatic color detection unit is provided for detecting a chromatic color picture element series in a main scanning line having a consecutive succession of chromatic color picture elements including edge picture elements. A replacing unit is provided for replacing the hue value of each picture element of the chromatic color picture element series with a median hue value, the hue value of a picture element located approximately at the center position of the detected chromatic color picture element series. The apparatus may further have an averaged hue value switching unit for switching the hue value of each of the picture elements with an average of the hue value of the picture elements and of other picture elements located in a nearby region.

Abstract: A holographic optical correlator includes a matched optical filter and control means therefor including a memory. In use searching for occurrences of an item to be found is carried out by the control means controlling the input and Fourier transform planes of the filter and a coherent light source for parallel optical processing of the memory content and the output plane of the filter provides information to the control means as to the location in the memory of occurrences of said item. The complexity of the detector means in the output plane can be reduced in comparison with previous proposals by moving the item to be sought step-wise across the input plane. The detector means then may comprise a single column of detectors which are required to detect a peak correlation signal within a predetermined range.