International Business Machines Corporation

Hiroyasu Takahashi

Naoyuki Nemoto

Shin Katoh

Akio Yamashita

Nobuyasu Itoh

Toshimichi Arima

Tomio Amano

Tetsunosuke Fujisaki

Richard G Casey

Jung-Kook Hong

Michitoshi Sumikawa

William David Modlin

Kottappuram Mohammedali Mohiuddin

Machiko Sato

To cut out a character and the like which are relatively darker than a background and are written with a pen and so on at a high speed, to use the cutting-out of a character and the like as pre-processing for character recognition, and to emphasize an object so as to hardly damage determinability of a character, a figure and the like to binarize the same. Attention is paid to that a character written with a pen and so on is relatively darker than a background, and a relative density of the character to a periphery thereof is obtained to binarize the same, thus emphasizing an object such as a character.

Image processing method, relative density detecting method and image processing apparatus

A character string extraction apparatus comprises: a connected component (CC) detector for detecting, in a binary image, connected components (CC) comprising black pixels; a character-sized connected component (CharCC) extraction unit for extracting character-sized connected components (CharCC) having an appropriate size from the detected connected components; a horizontal extension unit and a vertical extension unit for extending the extracted character-sized connected components in an assumed character string direction, and for reducing the character-sized connected components in a direction perpendicular to the assumed character string direction; long connected component (LongCC) extraction units and for connecting a plurality of the thus obtained connected components in the assumed character string direction, and for extracting a long connected component; and a character string selector for employing the extracted long connected component to determine a character string for image recognition.

Character string extraction and image processing methods and apparatus

An image processing apparatus comprises: a binarization unit for obtaining a binary image for the image entered by an image input unit ; a connected component detector for detecting the obtained connected components; a comparator for comparing the size of the detected connected components with a predetermined threshold size; a mesh image forming unit for dividing the image entered by the image input unit into mesh images having a predetermined size; a corresponding mesh image detector for detecting, from the mesh images, a mesh image that corresponds to a connected component that is determined by the comparator to occupy a range within the threshold size; a specific area extraction unit for extracting a specific area in accordance with the connection state of the corresponding mesh image that is detected; and an image recognition unit for recognizing an image that is located in the extracted specific area.

Area extraction method, address area extraction method, address area extraction apparatus, and image processing apparatus

Methods and apparatus are provided for removing ruled lines from a binary image containing character portions and ruled line portions, comprising the following steps. First, horizontal black runs in the bit-map image are detected and the detected black runs portions are stored as a run-length table comprising values of the horizontal start point and the length from the starting point of each black run for each vertical position. Then, based on the run-length table, black runs exceeding a predetermined threshold value are removed from the image to remove the ruled line comprising those black runs. Then, residual noise is eliminated from the image after the removal of the ruled lines. Then, in the image after the elimination of the residual noise, vertically disconnected components of a character are connected. Then, a portion of the character which is deleted by ruled line removal process is extracted from the image after the connection of vertically disconnected components of the character. Finally, the extracted portion of the character is interpolated in the image after the removal of the residual noise to restore the portion of the character removed by the ruled line removal process.

Method and apparatus for removing ruled lines

One or more contours of a picture element (PEL) array pattern are detected and, then, a strength of concavity/convexity on each contour (acuteness) and a direction of concavity/convexity are determined. Using an acuteness value of a contour point of interest, its curvature is classified into a group of strong curvatures or another group of weak curvatures such that the definitions of a direction of the concavity/convexity (contour direction) is appropriately switched from one to another for calculation. In a preferred aspect of this invention, using an acuteness value of a contour point of interest, its contour curvature is classified into either one of five curvature groups, comprising &ldquo;strong concavity,&rdquo; &ldquo;weak concavity,&rdquo; &ldquo;straight line/inflection point,&rdquo; &ldquo;weak convexity,&rdquo; and &ldquo;strong convexity.&rdquo; If its contour curvature is classified into either &ldquo;strong concavity&rdquo; or &ldquo;strong convexity,&rdquo; then calculation of a normal line direction will be made for deriving a contour direction. On the other hand, if its contour curvature is classified into &ldquo;straight line/inflection point,&rdquo; then calculation of a tangent line direction will be made. In so doing, it is possible to improve image recognition rates compared to those of the prior schemes utilizing conventional feature values.

Feature value extraction methods and apparatus for image recognition and storage medium for storing image analysis program

An object is to enable designation of character frames and recognition of characters even where a document does not have any page mark or reference mark nor does a scanner have a function for detecting an edge of the document. Also, to enable identification processing of a bitmap image in an accerelated manner by comparing bitmap images on the basis of a circumscribed rectangle, which is formed solely from horizontal line segments that are recognizable at high-speed. In a case of identifying a bitmap image comprising horizontal line segments such as character frames or grid lines on a document (as done in an OCR when it detects a document without page marks but with black character frames and recognizes characters), the horizontal line segments are extracted as a feature of the document, a circumscribed rectangle is formed within an area generated by the horizontal line segments such that it is used as information for identifying an estimated basis of character frame's positions and a class of the document. Applying this to an OCR, it becomes possible to identify even a document without page marks or reference marks thereon. Also, comparing the extracted horizontal line segments themselves with horizontal line segments of pre-registered document definition set to determine similarity between them, it becomes possible to identify a document in a more precise manner.

Apparatus and method of bitmap image processing, storage medium storing an image processing program

An automated optical character recognition method is provided for use in conjunction with a programmable digital processing device. The method inputs a sequence of values representing one or more characters in an array of characters to be optically recognized. The values define one or more dimensional characteristics of the characters. From the input values, a standard dimensional value is determined from a frequency distribution of a selected one of the character dimensional characteristics. For each of the input characters, a set of normalized values is determined from the standard dimensional value. The normalized values correspond to the one or more character dimensional characteristics. Optical character recognition is thereafter performed using the normalized values.

Speed and recognition enhancement for OCR using normalized height/width position

A mask process such as smoothing of a graphic or a character drawn into a bit pattern is performed by using only SHIFT, logical NOT, AND, OR operations of bit strings, rather than using conditional branch instructions in the form of &ldquo;IF THEN ELSE&rdquo; and &ldquo;SWITCH&rdquo; statements that lead to undesirable reduction of processing speed. This results in a substantial increase of the processing speed. Further, in order to process an image that is wider than a bus width or a register length at high speed, adjacent patterns of an original image, which are not necessarily used for a logical operation, are loaded onto left and right sides of a processing object, to form a processing unit for use in a logical operation.

Method and apparatus for processing an image, storage medium for storing an image processing program

The hybrid handwriting recognition method includes the steps of (a), in response to a handwriting input from a user, providing dynamic, time ordered stroke information; (b) determining a first list comprised of at least one probable character that the dynamic, time ordered stroke information is intended to represent; (c) converting the dynamic, time ordered stroke information to static stroke information; (d) determining a second list comprised of at least one probable character that the static stroke information represents; and (e) merging the first list and the second list to provide a third, unified list comprised of at least one element representing a most probable character that the dynamic, time ordered stroke information is intended to represent. The step of converting includes the steps of generating a static, bit-mapped representation of the dynamic stroke information, and generating one or more first stroke features based on contour directions of the bit-mapped stroke information. The first stroke features are applied as inputs to a plurality of neural network recognizers.

Hybrid on-line handwriting recognition and optical character recognition

Character normalization is accomplished using a scaling program. Data input to a scaling program are character images and character parameters (height of result buffer frame, character height, character width and character location), and parameters calculated for scaling (the size of a normalized portion to be output; and the top and bottom and the left and right hand spaces of a normalization pattern to be output). In the present invention, normalization is performed by adjusting the parameters, without especially changing the processing flow. A threshold value is provided in accordance with the size and the location of an input character in order to change a calculation method for parameters that are to be transmitted to the scaling program. Since the values at the boundary of the threshold value are continuous, no discontinuity appears in a resultant normalized image that covers the range from small characters to large characters, and seamless normalization is therefore accomplished.

Optical character recognition system and method using special

A system that allows a user to describe a tree or a set of trees in a table, and the system automatically generates a data structure corresponding to the tree or the set of trees in memory in accordance with the description. The system interprets a table with one or more columns wherein each of the node attributes given to a node are entered, allocates a memory area to each of the nodes, sets data on the attributes given to the node, and generates pointer data indicating the connections of the memory areas. The hierarchy of the nodes and the sequence of nodes linked immediately below the same node are determined on the basis of the sequence of rows corresponding to the nodes and levels assigned to the nodes in the table.

System and method for describing and creating a user defined arbitrary

Described is an image processing system which is operative to automatically determine a quadrilateral object such as a character frame, page mark, or position correction mark only with a mouse click. A field is automatically specified by displaying a scanned image of a form including a black frame on a display, clicking within a character frame at the left end for each recognition field, and clicking within a character frame at the right end of the same field. In this case, a field position/size determination program scans the image in the vertical and horizontal directions from the two clicked points to detect the inner wall of the black frame, and produces a histogram by establishing rectangles between two character frames to automatically detect the number of character frames in the field and the thickness of the black line between the character frames. In addition, for a form with a page mark or position correction mark, what is the mark can be automatically detected from the surrounding is black, and the height of the mark only by clicking the inside of the mark with a mouse.

Method and apparatus for recognizing a quadrilateral object contained in

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.-- RATE is calculated by using the minimum distance value to that group, and the second phase screening is performed.

Pattern recognition methods and apparatus to reduce processing time by

OCR processing involves: acquisition of images from image capturing devices; determination of location/size/style for each character; normalization; smoothing for large corners; smoothing by the deletion of individual protruding and extraneous dots; extraction of features; pattern matching; automatic error correction using a word dictionary; and output of recognition results. In the present invention, high speed software processes accomplish smoothing large corners and deletion of individual protruding and extraneous dots. In this software processing, the smoothing process is performed by shifting bit rows and using only the logical operators NOT, AND, and OR; not used are the often employed conventional conditional branch commands, such as the 'if then else' and 'switch' statements. This smoothing method (using only NOT, AND, and OR) contributes to a substantial reduction in the processing speed.

High speed outline smoothing method and apparatus including the

An apparatus for separating an image from a black frame quickly and precisely wherein corner coordinates for a character field are inferred by using a parameter relative to a character frame layout, and more precise corner coordinates are detected by pel distribution in a partial range. The positions of inner walls of a black frame are then inferred by using the detected corner coordinates and the parameter relative to the character frame, and a detection area is set. The positions of the inner walls are precisely detected by the pel distribution in the detection area, and smoothing process is performed on the detected values by employing the detected values of other adjacent black frames. Sequentially, white lines are drawn along the inner walls to separate the black frame from the character, and noise is removed. In consonance with a skew value, the detection area is divided to change the set range, and data separation is performed by using two white lines with a step difference between them. Through this processing, even when a skew value is large, the position of a black frame can be precisely determined, and since scanning is performed only in a partial range, fast processing can be provided.

Growing community of inventors

Yokohama, Japan

Hiroyasu Takahashi