Abstract: In an apparatus for analyzing a layout of a document, a character candidate element generator generates character candidate elements from black pixel linkage components of a document image. A horizontally oriented line rectangle generator sets a plurality of character candidate elements as a line candidate rectangle, among character candidate elements aligned in horizontal line orientation, when each amount of displacement of the set character candidate elements in a vertical orientation with respect to the horizontal line orientation, is smaller than or equal to a threshold value. A horizontally oriented paragraph-box generator sets a plurality of line candidate elements having approximately the same length as each other in the vertical orientation, as a paragraph candidate element.

Abstract: A document image tilt detection unit extracts a set of elements forming a character from a document image, extracts candidates for horizontal and vertical lines based on the character elements, estimates the reliability of each candidate for a line, and extracts a set of probable lines. Then, it estimates the tilt of the document image based on the arrangement of the character elements belonging to the line elements in the set of probable lines. A layout analysis unit extracts a set of lines in a fixed line direction from a set of basic elements forming the document image, and reciprocally performs the extraction of a column based on the correlation between lines and the extraction of a line based on the restrictions of the column. By providing the document image tilt detection unit and the layout analysis unit, the tilt of the document image containing both vertical and horizontal character lines can be detected, and lines and columns can be extracted at a high speed with high precision.

Abstract: The value of a reference feature vector calculated from a feature vector of an input pattern is transformed into a candidate category set, by a mapping described in a candidate table. Then, pattern recognition is performed using the candidate category set. By suitably setting the mapping, a high-speed process is performed while maintaining a recognition accuracy.

Abstract: After an original image, including half-tone dot meshed characters are inputted and a prescribed process is applied to the original image, black pixel connection patterns are extracted from a target process area. If the density of the extracted pattern exceeds prescribed density, it is judged that the pattern is half-tone dot meshed and both the respective average values and standard deviations of the vertical/horizontal sizes of the connection pattern are calculated. Then, respective threshold values are calculated based on both the respective average values and standard deviations, and half-tone dot patterns are eliminated. Furthermore, a pattern, the outline of which is long compared with an included black pixel, of all the remaining connection patterns is eliminated as an unnecessary pattern. Lastly, projections attached to a character are deleted and character extraction is terminated.

Abstract: A processing device performs region identification of an input image, and then performs an intra-region recognition process. The type code of each region and the individual code of a recognition result are then displayed, so that a user can modify both of the results of the region identification and the recognition process at one time. Furthermore, the processing device displays an original image close to the recognition result. If no correct answer exists among recognition candidates, code is added to the original image, and the original image with the code added is handled as a recognition result.

Abstract: A multilevel image into which a color or a back-and-white image is converted is input, and a slightly indistinct binary image generating unit generates a slightly indistinct binary image that includes a slightly indistinct line pattern and does not include background noise. Additionally, a shape-preserved binary image generating unit generates a binary image that preserves the shape of a line pattern and includes background noise. These images are ANDed for each pixel, so that a binary image that preserves the shape of the line pattern and does not include the background noise is generated.

Abstract: A key word is first and automatically extracted from a character string group to be recognized, and entered. Then, a character is recognized by segmenting an individual character from a character string image to be recognized, and a character string corresponding to the extracted/entered key word id extracted. Then, a word area delimited by a key word is extracted from the character string image, and a word is recognized. Furthermore, a word recognition result is verified, and a final character string recognition result is output.

Abstract: The present invention provides a character recognition method and recording medium for recognizing characters entered in a document or the like including preprint information. An object of the invention is to recognize characters in contact with or superposed on the preprint information without using any knowledge of the preprint information or a difference in gray level. An image in an area where the characters to be recognized are present is divided into line segments individually, wherein the image is obtained by reading the preprint information and the entry characters. A recognition image is created by changing a combination of the plurality of line segments divided. A recognition result with the reliability is memorized by making the character recognition for the created recognition image. And the recognition result having a greatest reliability is output by executing the character recognition for all the combinations while changing the combination of line segments successively.

Abstract: A three-dimensional curved-surface model can be estimated by using both two-dimensional outlines obtained from a piece of image photographed from the top and a restriction that the paper is rectangular. Then, only the three-dimensional distortion in the image can be corrected based on the obtained three-dimensional curved-surface model.

Abstract: A document image tilt detection unit extracts a set of elements forming a character from a document image, extracts candidates for horizontal and vertical lines based on the character elements, estimates the reliability of each candidate for a line, and extracts a set of probable lines. Then, it estimates the tilt of the document image based on the arrangement of the character elements belonging to the line elements in the set of probable lines. A layout analysis unit extracts a set of lines in a fixed line direction from a set of basic elements forming the document image, and reciprocally performs the extraction of a column based on the correlation between lines and the extraction of a line based on the restrictions of the column. By providing the document image tilt detection unit and the layout analysis unit, the tilt of the document image containing both vertical and horizontal character lines can be detected, and lines and columns can be extracted at a high speed with high precision.

Abstract: A rather expanded binary image and a rather blurry binary image are generated from a multiple-valued image. A ruled line candidate area is extracted from the rather expanded binary image, and the extracted ruled line candidate area is verified using the rather blurry binary image.

Abstract: A color image input from a document image inputting unit is converted into a gray-scale image by a brightness image extracting unit. The gray-scale image is then converted into an image having a higher resolution according to the resolution of the original gray-scale image. When this conversion is performed, subpixels are generated between the original pixels, and the values of the subpixels are obtained with an interpolation method. Furthermore, a threshold value for a binarization process is generated by using an original pixel value and a subpixel value. The characters included in the binarized image are recognized by a binary image recognizing unit, and a recognition result is output from a recognition result outputting unit.

Abstract: A character string extraction apparatus extracts an aggregate of basic components from a document image, such as a binary image, gray scale image, color image, etc., and judges whether each basic component is a character component using an inclusion relationship between the basic components. Then, the character string extraction apparatus extracts an aggregate of character components based on the judgment result and extracts a character string from the aggregate of character components.

Abstract: A document image tilt detection unit extracts a set of elements forming a character from a document image, extracts candidates for horizontal and vertical lines based on the character elements, estimates the reliability of each candidate for a line, and extracts a set of probable lines. Then, it estimates the tilt of the document image based on the arrangement of the character elements belonging to the line elements in the set of probable lines. A layout analysis unit extracts a set of lines in a fixed line direction from a set of basic elements forming the document image, and reciprocally performs the extraction of a column based on the correlation between lines and the extraction of a line based on the restrictions of the column. By providing the document image tilt detection unit and the layout analysis unit, the tilt of the document image containing both vertical and horizontal character lines can be detected, and lines and columns can be extracted at a high speed with high precision.

Abstract: The value of a reference feature vector calculated from a feature vector of an input pattern is transformed into a candidate category set, by a mapping described in a candidate table. Then, pattern recognition is performed using the candidate category set. By suitably setting the mapping, a high-speed process is performed while maintaining a recognition accuracy.

Abstract: A processing device performs region identification of an input image, and then performs an intra-region recognition process. The type code of each region and the individual code of a recognition result are then displayed, so that a user can modify both of the results of the region identification and the recognition process at one time. Furthermore, the processing device displays an original image close to the recognition result. If no correct answer exists among recognition candidates, code is added to the original image, and the original image with the code added is handled as a recognition result.

Abstract: A feature extracting unit extracts an input feature vector from an input character pattern. A learning unit computes orthonormal bases of a partial eigenspace of an original feature space from learning feature vectors of a plurality of learning character patterns, and stores the computation results in an orthonormal base storage unit for the partial eigenspace. The learning unit computes each of the dictionary selection feature vectors prescribed in the partial eigenspace corresponding to each of the object character patterns, and stores the computation results in a recognition dictionary unit. A feature selection unit computes an input selection feature vector obtained by projecting an input selection feature vector extracted by the feature extracting unit on the partial eigenspace using the above described orthonormal bases. A collating unit estimates a type of the input character pattern by collating the above described input selection feature vector with each of the dictionary selection feature vectors.

Abstract: A recognizing apparatus for estimating an environment description parameter, and substantially reducing a memory capacity required to represent a parameter space. The recognizing apparatus stores a restricted parameter space as a parameter space restricted by predetermined observation point information and based on an object model as shape information. The apparatus votes for a parameter subset consistent with the object model for each characteristic point obtained through environment observation in a restricted parameter space. The apparatus then outputs an estimated value for a environment description parameter according to the result of the voting for the restricted parameter space.

Abstract: A parallel computer includes a sequence of adjacent nodes, with each node including at least first and second processing elements and each processing element including a memory. The nodes are sequentially added or replaced in a node order and the processing elements are sequentially connected to form a single ring data path. First matrix data elements are sequentially assigned to adjacent processing elements along the single ring data path and second matrix data element groups are sequentially assigned to processing elements in the node order. The first matrix data elements are then accumulated with the second matrix data element groups as the first matrix data elements are rotated along the single ring data path. Accordingly, accumulation results are maintained within the memories of the lowest order nodes such that the nodes may added or replaced without relocation of data.

Abstract: A parallel data processing system uses a parallel learning method capable of having sufficient parallelness to shorten learning time and has plural data processing units, each connected to a data transfer unit and having a unit for holding execution parameters as are required for data processing, a unit for holding partial sample data which comprises at least part of the full sample data necessary for the required data processing, an adjustment value calculation unit which calculates, from partial sample data stored in the unit for holding partial sample data and from the execution parameters held in the execution parameter holding unit, adjustment amounts related to the execution parameters with regard to the partial sample data and, further, an accumulator which, when calculating the overall total of the execution parameter adjustment amounts with regard to the full sample data, accumulates the adjustment amounts related to the execution parameters with regard to the partial sample data at the data processi