Abstract: A method for detecting stellate lesions in digitized mammographic image data includes the steps of prestoring a plurality of reference images, calculating a plurality of features for each of the pixels of the reference images, and creating a binary decision tree from features of randomly sampled pixels from each of the reference images. Once the binary decision tree has been created, a plurality of features, preferably including an ALOE feature (analysis of local oriented edges), are calculated for each of the pixels of the digitized mammographic data. Each of these plurality of features of each pixel are input into the binary decision tree and a probability is determined, for each of the pixels, corresponding to the likelihood of the presence of a stellate lesion, to create a probability image. Finally, the probability image is spatially filtered to enforce local consensus among neighboring pixels and the spatially filtered image is output.

Abstract: A method for indexing a plurality of handwritten objects is provided. A B-tree data structure of order m is generated, where m is an integer. The B-tree has a plurality of nodes divided into a plurality of levels ordinally numbered 0.sup.th through n.sup.th. Each node in the 0.sup.th level is a leaf. Each node in the 1.sup.th level has at least m/2 leaves as children. Each one of the handwritten objects is assigned to a respective leaf. A respectively different hidden Markov model (HMM) is associated with each respective child of each of the nodes in the 1.sup.th to n.sup.th levels. Each one of the nodes in the 1.sup.th to n.sup.th levels contains the respective HMM associated with the child of the one node. Each HMM in each one of the nodes in the 1.sup.th level is trained to accept the handwritten object of the respective leaf that is a child of the one node. Each HMM associated with any of the nodes in the 2.sup.th through n.sup.

Abstract: A recognition system that includes a development subsystem and production subsystem is disclosed. The development subsystem includes a user interface that enables a developer to describe the objects to be recognized and their relationships in a recognition tree; a data store system for storing the descriptions; and a technique for automatically generating a decision tree from the object descriptions. The production subsystem includes an engine for executing the tests on the input data at each node of the decision tree, traversing the decision tree, building a collection of identified objects, and finding the leaf of the decision tree that is the answer for a given input.

Abstract: A method is disclosed for matching input data representing a continuous combination of input objects to a plurality of objects in a trie database structure. This data structure has a plurality of nodes partitioned into a plurality of levels. Each node in the Trie includes a plurality of elements where each element corresponds to a respective one of the component objects. In addition, a hidden Markov model corresponding to the component object is associated with the element in the database. According to the method, the input object is applied to each of the hidden Markov models associated with the respective plurality of elements of a node to generate a respective plurality of acceptance values. The element which generates the largest acceptance value is identified with a segment of the input data. The component object for this element is recorded and the identified segment is deleted from the input data string.

Abstract: A top-down technique for character text recognition of an image comprises a left-to-right analysis of each image line. A current image portion is selected. Possible text prefixes are selected from a dictionary. The upper and lower text contours of the text prefixes are compared with a bitmap of the current image portion. A distance value is generated, indicating the quality of the comparison. The prefixes are then added to an agenda of prefixes. Based on the distance value, corresponding to the similarity of the upper shapes and lower shapes of the possible prefix to the bitmap of the image portion, a list of the text prefixes generating the best distance values is selected from the agenda. From the selected list, a new list of extended text prefixes is obtained from the dictionary and added to the agenda. The process is repeated until the current image portion ends.

Abstract: Pattern recognition system which provides an indication of the confidence with which a candidate is selected for an unknown pattern. The pattern recognition apparatus includes an image data input device, a host for segmenting the image data into unknown patterns, and a character recognition device for providing a candidate for each unknown pattern. The character recognition device includes a confidence level indicator for providing a confidence level indication. In one aspect, the confidence level indication is determined based on the proximity of an unknown pattern to a known pattern. In another aspect, the confidence level indication is determined based on the consistency with which the unknown pattern is recognized using different recognition functions. In yet another aspect, the confidence level indication is determined by ensuring that a candidate is not provided unless the candidate is closer than a predetermined distance from a known pattern.

Abstract: A dividing step (a) divides into segments at least one character line forming input characters to be read. A partial-figure forming step (b) forms partial figures by combining segments from among the thus obtained segments. A character reading step (c) attempts to read each of said partial figures as a character. A network forming step (d) forms a network wherein partial figures, among the partial figures, which have been read in said character-reading step (c) are used as nodes and said nodes are connected with one another by links, and wherein said links respectively have appropriate weights. An optimum-path selecting step (e) selects an optimum path from among paths existing in said network so that said optimum path comprises nodes, from among said nodes, which respectively correspond to said input characters.

Abstract: Video data compression techniques reduce necessary storage size and communication channel bandwidth while maintaining acceptable fidelity. Vector quantization provides better overall data compression performance by coding vectors instead of scalars. The search algorithm and VLSI architecture for implementing it is herein disclosed, and such a search algorithm is useful for real-time image processing. The architecture employs a single processing element and external memory for storing the N constant value hyperplanes used in the search, where N is the number of codevectors. The design does not perform any multiplication operation using the constant value hyperplane tree search, since the tree search method is independent of any L.sub.q metric for q between one and infinity. Memory used by the design is significantly less than memory employed in existing architecture.

Abstract: A document classifying system is formed by a document data classifying system, a document classifying function building system, a sample data storage apparatus and a keyword storage apparatus. The document data classifying system inputs document data and defines a classification to which the document data belongs. The document classifying function building system is operatively connected to the document data classifying system and automatically builds a classification decision tree in the document data classifying system. A sample data storage apparatus is operatively connected to the document classifying function building system and stores sample data formed by a set comprised of the document data and the classification to which the document data belongs. The keyword storage apparatus is operatively connected to the document classifying function building system and stores keywords extracted thereby.

Abstract: Coding and decoding of digital images by transformation of pixel blocks reduce the data bit rate. Transformation of blocks leads to discontinuities in brightness level for pixels bordering on block boundaries. These discontinuities are detected for each given bordering pixel by a change of sign of two level gradients respectively computed with the levels of two pixels adjacent to the bordering pixel. A brightness level of the bordering pixel is then corrected by a discontinuity correction factor evaluated from a discontinuity divergence equal to the absolute value of the difference between the brightness level of the bordering pixel and an arithmetical average of brightness levels of the two adjacent pixels, as a function of a quantization pitch associated with a block to which belongs the bordering pixel. The block is then corrected in its entirety to attenuate the differences of average brightness levels between two adjacent blocks generated by the coding and decoding.

Abstract: A post-processing method for an optical character recognition (OCR) method for combining different OCR engines to identify and resolve characters and attributes of the characters that are erroneously recognized by multiple optical character recognition engines. The characters can originate from many different types of character environments. OCR engine outputs are synchronized in order to detect matches and mismatches between said OCR engine outputs by using synchronization heuristics. The mismatches are resolved using resolution heuristics and neural networks. The resolution heuristics and neural networks are based on observing many different conventional OCR engines in different character environments to find what specific OCR engine correctly identifies a certain character having particular attributes. The results are encoded into the resolution heuristics and neural networks to create an optimal OCR post-processing solution.