Abstract: Provided is a method and apparatus of processing an image using filters. The method and apparatus receives an input pixel and a pixel window associated with the input pixel from the image, classifies the input pixel using the pixel window into a range of classes identifying pixels suitable for various degrees of smoothing and sharpening operations, receives parameter independently set for sharpening and smoothing the image, and selects a filter for processing the input pixel based upon the pixel classification and the parameter settings for sharpness and smoothness.

Abstract: It is an object to print an image as it is inherently formed. Objects constructing the image are separated to character train objects in which there is no need to consider an overlap and the other objects by a character train separator. A character train converter converts the separated character train objects in which there is no need to consider the overlap into character code information and transmits to a printer. The other objects are converted into image information by a drawer and transmitted to the printer. The printer synthesizes a glyph formed by a glyph generator from the character code information onto the received image information and outputs.

Abstract: A method of conversion of a quadtree representation of image data into a corresponding representation of edges of regions within the image is disclosed. In one configuration the method comprises recursively processing (70) each quadrant using a number of steps. A first step examines if the quadrant is of a first particular uniform type (74), and if so, the method forms a series of lists (76) defining the boundary structure of the quadrant. A second step examines if the quadrant is of a second particular uniform type (80), and if so, the method forms a series of empty lists (82). A third step is applied where a quadrant is of an intermediate type containing data values of the first particular uniform type and the second particular uniform type, and the method divides the quadrant into a series of sub-quadrants (84,88) and recursively (86) applies the first, second and third steps to each of the sub-quadrants in accordance with the quadtree representation.

Abstract: Described herein is a technology for semantically classifying areas of an image (and/or the images themselves) as one of a number of multiple discriminating categories. More particularly, the technology employs one or more hierarchical, probabilistic techniques for performing such classification. Such technology is particularly useful in fields of image classification and image retrieval. The architecture of such technology employs multiple hierarchical layers. The architecture is based on modeling class likelihoods at each of such layers separately and then combining these to form an overall estimate of the posterior, conditioned on the data. The task of combining the estimated class likelihoods at each layer is made more efficient by assuming statistical independence between layers. This abstract itself is not intended to limit the scope of this patent. The scope of the present invention is pointed out in the appending claims.

Abstract: A method for region analysis of a document image applied to region analysis system of a document image, the method includes the steps of: a) analyzing a connected component though a reduced documentimage; b) classifying the connected component by generating a tree according to analysis result of the connected component; c) grouping text components from the classified connected component according to a spatial connection; and d) refining a text block by repeating segmentation and merge of the connected component after the grouping.

Abstract: An image analysis system detects objects and patterns within a digital image. The system includes a synaptic link generator, a knowledge element (knixel) identifier, a theme identifier, and a corresponding theme/knixel associations knowledge base. The synaptic link generator includes a determinant extractor for computing a set of determinants for a plurality of pixels within the digital image. The synaptic link generator uses the determinants to generate a synaptic link for each pixel having a set of determinants and a knowledge element identifier is stored at an address corresponding to each synaptic link. The theme identifier evaluates associative links in a theme knowledge base to determine whether the knowledge elements or other themes identified as being present in the digital image conform to a logical relationship defined by the associative links.

Abstract: A method and apparatus for optical character recognition particularly suitable for cursive and scripted text in one or more of several languages follows the tracings of the script and encodes them as a sequence of directional vectors. It reads a preprocessed word or sub-word of interconnected characters as a unit and the characters are accepted in a specific language only if all characters in a unit have been recognized by testing against a first set of language-specific rules without leaving a remainder of any vectors in the unit. If there are vectors remaining unused, it moves a character marker to utilize more or fewer vectors for the current sub-word in order to obtain recognition. If the vectors do not form a character in a first language, the invention consults a second set of language-specific rules and follows similar steps to obtain recognition of one or more characters against the second set of language-specific rules.

Abstract: A quad-tree embedded image coding technique is used in combination with a bit-plane encoding technique to provide an efficient and low complexity embedded image coding system. A simple quad-tree method identifies coefficients as significant, insignificant, or refinement at each successive quantization level. The quad-tree technique is used instead of the zero-tree or hierarchical tree used in previous encoders.

Abstract: A method and apparatus is disclosed for generating a decision tree classifier with oblique hyperplanes from a training set of records. The method iteratively comprises the steps of: initializing a set of vectors to the numeric attribute axes; creating a decision tree classifier using hyperplanes orthogonal to the set of vectors; checking if the iteration stopping criteria has been reached; computing a new set of vectors if the iteration proceeds; and choosing the best decision tree when the iteration is stopped. The vectors used are not restricted to the attribute axes and hence oblique hyperplanes are allowed to split nodes in the generated decision tree. The computation of the new vector set uses the decision tree produced in the latest iteration. The leaf nodes of this tree are considered pair-wise to compute the new vector set for use in the next iteration. The iterative process produces a set of decision trees from which the best one is chosen as the final result of the method.

Abstract: The present invention relates to an information processing method for recognizing a pattern of an image, character, speech or the like. According to the method, a feature space in which a point set given as learning patterns is present is divided to form a classification tree on the basis of the learning patterns. The method comprises the linear combination feature amount generation step of generating a new feature amount by a linear combination of the feature amounts of the learning patterns; the hierarchization pre-preprocessing step of hierarchizing, in advance, the new feature amount generated in the linear combination feature amount generation step; and the classification tree generation step of generating a classification tree on the basis of the learning patterns hierarchized in the hierarchization pre-processing step.

Abstract: To identify an image characteristic, a tree structured code having coding representing each of a plurality of training images is stored in a memory. Coding representing a sample image, not included in the plurality of training images, is compared to the tree structured code to identify the training image coding in the structured tree code closest to the sample image coding. A characteristic of the sample image is identified based upon the training image represented by the closest training image coding.

Abstract: Systems and methods for reordering unconstrained handwriting data using both spatial and temporal interrelationships prior to recognition, and for spatially organizing and formatting machine recognized transcription results. The present invention allows a machine recognizer to generate and present a full and accurate transcription of unconstrained handwriting in its correct spatial context such that the transcription output can appear to “mirror” the corresponding handwriting.

Abstract: A method represents a shape of an object in an image. Portions of the image interior to a boundary of the object are ordered. A medial axis for the ordered portions of the image is derived. The medial axis is segmented to produce an ordered tree. The ordered tree is transformed to a partial ordered tree to represent the shape of the object.

Abstract: An object data processing apparatus for decoding N pieces of coded data (N=positive integer) obtained by compressively coding N pieces of object data which constitute individual data to be recorded or transmitted and have a hierarchical structure, for each object data. This apparatus includes hierarchical information extraction means for extracting hierarchical information showing the hierarchical relationship of the N pieces of object data, according to the coded data; and table creation means for creating, according to the hierarchical information, an object table on which the respective object data are correlated with coded data of the respective object data. Therefore, the apparatus can perform extraction, selection, or retrieval of coded data corresponding to a specific object at high speed, and this enables the user to edit or replace the object data in short time with high controllability.

Abstract: That is, a hierarchical coding of an image data is generated. An image data of a second hierarchy is formed, having a number of pixels which is smaller than that of an image data of a first hierarchy. The image data of the second hierarchy is corrected to generate a corrected data. The image data of the first hierarchy is predicted in accordance with the corrected data and a predicted data of the first hierarchy, having a plurality of predicted pixels, is generated. Predictive error of the predicted data of the first hierarchy is calculated with respect to the image data of the first hierarchy. The suitability of the corrected data is determined in accordance with the predicted error. A difference data, between the image data of the first hierarchy and the predicted data of the first hierarchy, is generated in accordance with the determined suitability. The difference data has a plurality of difference values. The corrected data and the difference data are output as the coded image data.

Abstract: An adaptive OCR technique for character classification and recognition without the input and use of ground truth derived from the image itself. A set of so-called stop words are employed for classifying symbols, e.g., characters, from any image. The stop words are identified independent of any particular image and are used for classification purposes across any set of images of the same language, e.g., English. Advantageously, an adaptive OCR method is realized without the requirement of the selection and inputting of ground truth from each individual image to be recognized.

Abstract: A classification tree which allows direct recoginition of an input pattern such as image or sound without extra processing such as pre-processing of unprocessed pattern data having high order characteristic variables is prepared. Information processing method and apparatus conduct hierarchical pre-processing for hierarchically pre-processing a learning pattern, prepares a classification tree based on the learning pattern processed by the hierarchical pre-processing and conducts the recognition by using the classification tree.

Abstract: A method of using a Chinese dictionary, including the steps of (a) selecting a stroke type of a first stroke of a principal root in a desired Chinese Character from among a corresponding stroke group found in a root table, the stroke group being a horizontal stroke, a vertical stroke and a slant stroke, the root table containing a root for the desired Chinese character together with a page where the desired Chinese character is found in the Chinese dictionary, (b) identifying the page from the root table that is associated with the selected stroke type of the first stroke, (c) selecting a stroke type of a first stroke in the secondary root from among the corresponding stroke group, (d) finding on the page a list of Chinese characters associated with the selected stroke type of the first stroke in the secondary root, (e) selecting stroke types of the next one or two strokes in sequence in the secondary root from among the corresponding stroke group, (f) finding a subsidiary list of Chinese characters from the

Abstract: Pattern matching means 3 matches categories of a reference pattern stored in reference pattern memory means 2 and input samples of an input pattern produced by input pattern generating means 1. Data statistics estimating means 6 calculates the numbers of input samples in individual nodes of a tree scheme reference pattern stored in the tree scheme reference pattern memory means 4. Node selecting means 6 selects nodes used for adaptation according to the input sample numbers calculated by the data statistics estimating means 5. Adaptation parameter generating means 7 calculates an adaptation parameter in the Nodes selected by the node selecting means 6. Reference pattern generating means 8 produces an adapted reference pattern using the calculated adaptation parameter.

Abstract: A method for automatically classifying test images based on their similares with a dictionary of example target and non-target images. The method operates by receiving a test image and then initializing variables for an iteration count and for the linear expansion of the test image. The test image is then projected onto each one of the target and non-target images in the dictionary, wherein a maximum scaling coefficient is selected for each iteration. A residue is then generated, and the linear expansion of the test image is increased until a predetermined number of iterations have been performed. Once this predetermined number of iterations have been performed, the sum of the scaling coefficients belonging to the target examples in the dictionary is compared to the sum of the scaling coefficients belonging to the non-target examples in the dictionary to determine whether the image is a target signal or a non-target signal.

Abstract: A method and apparatus are disclosed for determining a direction in which a subject's head is facing. The present invention has particular relevance to monitoring driver conditions in automobiles. By monitoring head positions and motions, the system can estimate driver awareness. Signals or alarms can be activated if the driver does not appear to be aware of driving conditions. The present invention provides a rapid, low cost device for determining pose position, which would be useful in automobiles. A large number of images of various subjects in different poses are stored a memory. The poses are then organized according to tree-structured vector quantization to allow comparison with a sample image of an actual driver. The tree structure is traversed to find a leaf node constituting one of the stored images which is closest to the sample image. The pose of the closest image is outputted as the pose of the sample image.

Abstract: A free-lying cell classifier. An automated microscope system comprising a computer and high speed processing field of view processors identifies free-lying cells. An image of a biological specimen is obtained and the image is segmented to create a set of binary masks. The binary masks are used by a feature calculator to compute the features that characterize objects of interest including free-lying cells, artifacts and other biological objects. The objects are classified to identify their type, their normality or abnormality or their identification as an artifact. The results are summarized and reported. A stain evaluation of the slide is performed as well as a typicality evaluation. The robustness of the measurement is also quantified as a classification confidence value. The free-lying cell evaluation is used by an automated cytology system to classify a biological specimen slide.

Abstract: The speed and accuracy of a computer implemented handwriting recognition system is enhanced by several innovations, including integrated segmentation and context processing. The recognition processing occurs while the user is providing ink data. The system quickly reaches the recognition result once all of the input is received. More than one result may be returned by the system.

Abstract: A fast way for determining the best subset test for a nominal attribute in a decision tree. When a nominal attribute has n distinct values, the prior art requires computing the impurity functions on each of the 2.sup.n-1 -1 possible subset partitioning of the n values and finding the minimum case among them. This invention guarantees the minimum impurity test on the attribute by computing only (n-1) impurity function computations. This reduction of computational complexity makes it practically possible to find the true best tests for many real data mining application, where a binary decision tree is used as the classification model.

Abstract: A method of image processing for tracking a subject in a sequence of image frames using predefined criteria for classifying image regions as potentially being part of the tracked subject includes steps for obtaining, for a first image frame, a record of an image region previously designated as containing the tracked subject, classifying image regions of a second image frame in accordance with the defined criteria to thereby generate classified image regions, and updating the designated image region associated with the second image frame with reference to both the classified image regions in the second image frame and the previously designated image region in the first image frame. Tracking the subject though a sequence of image frames permits processing the portion of the image frames corresponding to the subject at a relatively high resolution while processing the background at a relatively low resolution so as to maximize image processing for regions containing the subject.

Abstract: A new method is presented for efficient and reliable object recognition within noisy, cluttered, and occluded range images. An image is captured and quantized into cells having characteristics. A tree data structure is provided wherein each node of the tree is associated with a cell. The tree is traversed according to the characteristics of each associated cell using a best-first algorithm. When a leaf node is reached, potential object poses, types and locations associated with the leaf node are stored. Once sufficient information is stored, each potential recognized object is verified using a more reliable algorithm.

Abstract: In a decision-forest classifier in accordance with the invention, a decision forest including multiple decision trees is used to classify "seen" training data and "unseen" data. Each individual tree performs an initial classification based on randomly selected subsets of the data. The classification outcomes by the individual trees are combined using a discriminant process in the decision-forest classier to render the ultimate classification decision.

Abstract: A page analysis system which analyzes image data of a document page includes analysis of image data of one of a plurality of portions of the page, identification of a block of image data which has an association to another separate block of image data, and determination of whether or not picture information in the image data is half-tone line-drawing, joint-line, or unknown.

Abstract: A method and system are provided for distinguishing objects. Signals representative of the objects are input. From the signals, measurements of multiple types are acquired. At least one of the measurement types is continuously variable. In response to the measurements, respective amounts of mutual information of the measurement types are determined. In response to the respective amounts of mutual information, at least one measurement type is selected for evaluation in distinguishing the objects.

Abstract: This relates to a classification generation method whereby, in order to efficiently and accurately recognize a pattern having a large number of characteristics, a pattern classification tree is generated, with which a macro structural characteristic of a pattern is appropriately reflected and a competitive relationship between categories is adequately reflected, and to a method for recognizing an input pattern by using the generated classification tree. When an input pattern is formed using strokes, a training stroke is divided into a plurality of segments, and vector quantization is performed for the strokes in the segments. Among the quantized strokes in the segments, adjacent stroke sets are synthesized to repetitively generate upper rank stroke vectors. A stroke vector for which a predetermined entropy function is maximized is selected from the upper rank stroke vectors in a layered stroke vector series, and development is performed extending down into the lower rank stroke vector sets.

Abstract: The present invention related to a coding method suited to obtain at a high speed a code number determined according to a sequence of arrangement of a plurality of texts which are arranged in a tree structure and output the code number and a semiconductor memory for implementing the coding method wherein a node number is assigned to each of nodes of the tree structure and a node to which the retrieval should advance is obtained in accordance with both a text entered and a node number of a node where the retrieval currently stays.

Abstract: The detection of a nearest neighbor is calculated at high speed and with high accuracy. Feature space grouping is performed, with one, or two or more features being assigned to each group. A check is performed to determine whether or not a calculated distance value that is acquired before the limits for a group is reached has exceeded the threshold value THR.sub.-- MIX for each group. The threshold value that is set for THR.sub.-- MIX is the average of the threshold value that is determined based on the cumulative quantity of the features that is acquired and the threshold value that is based on the number of dimensions. When the distance value for each group is greater than the threshold value THR.sub.-- MIX, first phase screening is performed to exclude, from the succeeding distance calculations, the prototype that provides that distance value. Then, for a group for which the first phase screening is performed, the threshold value THR.sub.

Abstract: In a decision-forest classifier in accordance with the invention, a decision forest including multiple decision trees is used to classify "seen" training data and "unseen" data. Each individual tree performs an initial classification based on randomly selected subsets of the data. The classification outcomes by the individual trees are combined using a discriminant process in the decision-forest classier to render the ultimate classification decision.

Abstract: A method and apparatus to recognize hand-printed characters represented as an electronic ink stream. A first filtering box (300) is defined around a center of a first output point (324) in the electronic ink stream and the first output point is added to a filtered ink stream. A second output point (328) is added to the filtered ink stream only when the electronic ink stream moves outside the filtering box (300). When the second output point (328) is added, the filtering box is re-centered on the second output point (328) only in a dimension that is exceeded after adding the second output point (328) to the filtered ink stream to create a second filtering box (310). Moreover, extraction points (630, 640, 650 and 660) are provided approximately equally spaced across a finite set of points of electronic ink to create a compact representation of a finite set of points of electronic ink.

Abstract: A method of recognizing speech input selectively creates and maintains grammar representations of the speech input in essentially real time. Speech input frames are received by a speech recognition system. Grammar representations are created for each speech frame and a probability score is derived for the representations indicating the probability of the accuracy of the representations to the speech input. Representations having a probability score below a predetermined threshold are not maintained. Those grammar representations having probability scores above the predetermined threshold are maintained. As more speech frames are received by the system, additional grammar representations are created and the probability scores are updated. When the entire speech input has been received, the chain of grammar representations having the highest probability score is identified as the speech input.

Abstract: A method and system for simultaneously recognizing contextually related images is disclosed. The image of two separate fields is captured to form two captured data images such as a word and numerical amount. Each captured image is cut to form a segmentation graph based on the cuts. The shortest path in each segmentation graph is found wherein the additive length corresponds to a score and is associated with each directed arc of the segmentation graph. The segmentation graphs are combined into a joint segmentation graph and the highest scoring mutually consistent interpretations are found.

Abstract: A method and system for automatically modifying an original transcription produced as the output of a recognition operation produces a second, modified transcription, such as, for example, automatically correcting an errorful transcription produced by an OCR operation. The invention uses information in an input text image of character images and in an original transcription associated with the input text image to modify aspects of a formal image source model that models as a grammar the spatial image structure of a set of text images. A recognition operation is then performed on the input text image using the modified formal image source model to produce a second, modified transcription. When the original transcription is errorful, the second transcription is a corrected transcription. Several aspects of the formal image source model may be modified; in particular, character templates to be used in the recognition operation are trained in the font of the glyphs occurring in the input text image.

Abstract: A multi-stage, preprocessing technique increases the quality of a monochrome or colored mixed-mode document printed on a binary-type printer. The document is preferably apportioned into a plurality of small tiles, each of which provides information to the inventive technique for eventually defining the modes as either text or images. The multi-stage technique comprises a first stage for normalizing a non-ideal histogram of the mixed-mode document to a histogram having an ideal distribution of selected pixel values. A second stage classifies each tile as either text, image or background, while a third stage collates groups of tiles in both vertical and horizontal directions to remove variances in the document. Those portions of the document classified as images are then halftone processed, while the text areas are threshold processed to thereby significantly improve the reproduction quality of the printed document.

Abstract: A method for providing an effective completion of characters required in inputting a partial character string.

Abstract: Multi-dimensional data contained in very large databases is efficiently and accurately clustered to determine patterns therein and extract useful information from such patterns. Conventional computer processors may be used which have limited memory capacity and conventional operating speed, allowing massive data sets to be processed in a reasonable time and with reasonable computer resources. The clustering process is organized using a clustering feature tree structure wherein each clustering feature comprises the number of data points in the cluster, the linear sum of the data points in the cluster, and the square sum of the data points in the cluster. A dense region of data points is treated collectively as a single cluster, and points in sparsely occupied regions can be treated as outliers and removed from the clustering feature tree.

Abstract: A system for editing the hierarchical tree structure which is created by a block selection system to correspond to a block template which represents a document image, wherein the hierarchical tree structure includes a plurality of nodes, each of which represents a block of document image data in the block template of a document image and contains document feature data defining features of the block of image data. The system operates to download from memory the hierarchical tree structure, generate and display a block template representing a document image corresponding to the hierarchical tree structure in memory, select a block of document image data to be edited in the displayed block template, edit a feature of the selected block of image data and update the document feature data in a node corresponding to the selected block of image data.

Abstract: A method and apparatus for consolidation of equivalence classes of scanned symbols as may be used for image data compression. The present invention performs run-length symbol extraction and classifies symbols based on both horizontal and vertical run length information. An equivalence class is represented by an exemplar. Feature-based classification criteria for matching an exemplar is defined by a corresponding exemplar template. The feature-based classification criteria all use quantities that can be readily computed from the run endpoints. Reducing the number of equivalence classes is achieved through a process called equivalence class consolidation. Equivalence class consolidation utilizes the symbol classifier to identify matched exemplars indicating equivalence classes which may be merged. For a consolidated equivalence class, the exemplar matching the most symbols is selected as the representative for the class.

Abstract: Image processing apparatus for segmenting an input image into image portions each containing a single character, the apparatus comprising identification logic for identifying connected components in the input image; classification logic, including a neural network, for determining into which of a number of predefined classes a connected component falls, at least one of said classes indicating that the connected component is most likely to be a single character; merging logic and splitting logic for merging and splitting the connected components. The merging and splitting logic and the classification logic are arranged to operate so that the connected components are iteratively merged and/or split and the resulting split and/or merged connected components reclassified by the classification logic until an image segmentation is achieved which meets a predefined criterion.

Abstract: A method and apparatus for classification of scanned symbol into equivalence classes as may be used for image data compression. The present invention performs run-length symbol extraction and classifies symbols based on both horizontal and vertical run length information. An equivalence class is represented by an exemplar. Feature-based classification criteria for matching an exemplar is defined by a corresponding exemplar template. The feature-based classification criteria all use quantities that can be readily computed from the run endpoints. Reducing the number of equivalence classes is achieved through a process called equivalence class consolidation. Equivalence class consolidation utilizes the symbol classifier to identify matched exemplars indicating equivalence classes which may be merged. For a consolidated equivalence class, the exemplar matching the most symbols is selected as the representative for the class.

Abstract: A method of searching for one of a plurality of objects that matches an input sequence of handwritten objects is provided. The objects are modeled by concatenating members of a set of component objects. A Trie structure representing the plurality of objects is generated. Component objects of each object are assigned to the elements of respective nodes of the Trie structure. A respective hidden Markov model (HMM) is associated with each element of each non-leaf node. The HMMs represent the respective component object of the element. A maximum probability of any HMM accepting any of the set of component objects is estimated. The root node of the Trie structure is selected. A plurality of elements of the selected node are selected. A plurality of segments of the input sequence are applied to respective HMMs associated with the selected elements to generate respective acceptance values.

Abstract: An arrangement and method for detecting sequential processing effects on devices to be manufactured in a manufacturing process extracts data regarding responses of the devices to a process step in the manufacturing process and data regarding a processing sequence of the devices in that process step. The extracted data is refined before analysis and control chart rules are then applied to the refined data. These control chart rules detect whether there are any unusual processing effects caused by the sequence of processing of the devices in any one of the individual processing steps. Application of control chart rules to the refined data allows an automatic determination of whether there are any rule violations. One or more control charts which have a rule violation are automatically generated when it is determined that there is a rule violation. Process engineers may then use the automatically generated charts to direct their efforts at improving the manufacturing process.

Abstract: A method and system are provided for distinguishing objects. Signals representative of the objects are input (63). From the signals, measurements of multiple types are acquired (64). At least one of the measurement types is continuously variable. In response to the measurements, respective amounts of mutual information of the measurement types are determined (70). In response to the respective amounts of mutual information, at least one measurement type is selected for evaluation in distinguishing the objects (72).

Abstract: A method of recognizing speech input selectively creates and maintains grammar representations of the speech input in essentially real time. Speech input frames received by a speech recognition system. Grammar representations are created for each speech frame and a probability score is derived for the representations indicating the probability of the accuracy of the representations to the speech input. Representations having a probability score below a predetermined threshold are not maintained. Those grammar representations having probability scores above predetermined threshold are maintained. As more speech frames are received by the system, additional grammar representations are created and the probability scores are updated. When the entire speech input has been received, the chain of grammar representations having the highest probability score is identified as the speech input.

Abstract: In a character recognition system or the like, method and apparatus for selecting blocks of pixels from pixel image data so as to permit identification and grouping of similarly-typed pixels, such as text-type pixels and non-text-type pixels. Pixel image data is inputted and, if the pixel image data is not binary image data then the pixel image data is converted into binary pixel image data. Blocks of pixel image data are selected by outlining contours of connected components in the pixel image data, determining whether the outlined connected components include text unit or non-text units based on the size of the outlined connected components, selectively connecting text units widthwisely to form text lines based on proximity of adjacent text units, and selectively connecting text lines vertically to form text blocks based on proximity of adjacent text lines and on the position of non-text units between text lines. A hierarchical tree is formed based on the outlined connected components.

Abstract: A character recognition system or the like in which character identities are stored in accordance with a hierarchical order established during processing to separate text image areas from non-text image areas. To separate text image areas from non-text image areas, blocks of pixels are selected from pixel image data by outlining contours of connected components in the pixel image data, determining whether the outlined connected components include text units or non-text units, selectively connecting text units widthwisely to form text lines, and selectively connecting text lines vertically to form text blocks. After blocks of pixels have been so selected, text blocks are segmented into lines of pixel image data, and characters are cut from the lines of pixel image data so obtained. If desired, the characters may be cut by a two-step cutting process in which non-touching and non-overlapping characters are first cut out, and touching characters are then cut out.