Browse Prior Art Database

Character Recognition Apparatus

IP.com Disclosure Number: IPCOM000059976D
Original Publication Date: 1986-Feb-01
Included in the Prior Art Database: 2005-Mar-08
Document File: 3 page(s) / 67K

Publishing Venue

IBM

Related People

Nakamura, Y: AUTHOR [+2]

Abstract

This article describes a character recognition apparatus for recognizing multifont characters which may be either printed or handwritten. The apparatus scans character images, extracts predetermined features from each of the scanned images, performs preliminary classification and discrimination based on the extracted features, and then determines a single category or character by applying a tournament scheme to a plurality of categories remaining after the discrimination step. Fig. 1 shows the present apparatus which is wholly controlled by controller (CTL) 10. Scanner 12 scans character images to be recognized and digitizes them. The digitized images are stored as input patterns in random-access memory (RAM) 14 with each pattern stored in, for example, a 32 x 32 area.

This text was extracted from a PDF file.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 53% of the total text.

Page 1 of 3

Character Recognition Apparatus

This article describes a character recognition apparatus for recognizing multifont characters which may be either printed or handwritten. The apparatus scans character images, extracts predetermined features from each of the scanned images, performs preliminary classification and discrimination based on the extracted features, and then determines a single category or character by applying a tournament scheme to a plurality of categories remaining after the discrimination step. Fig. 1 shows the present apparatus which is wholly controlled by controller (CTL) 10. Scanner 12 scans character images to be recognized and digitizes them. The digitized images are stored as input patterns in random- access memory (RAM) 14 with each pattern stored in, for example, a 32 x 32 area. Feature extractor 16 extracts predetermined features from each of the input patterns stored in RAM 14 and calculates feature values in two steps. In the first step, feature extractor 16 applies a 2 x 2 mask for each pixel (picture element) in an input pattern to extract local con tour features. Fig. 2 shows sixteen possible local patterns detected by the mask together with their assigned local directions and feature values. The feature values are accumulated for each local direction and for each region, called look region, shown in Fig. 3. Fig. 3 shows that the input pattern is divided into six regions horizontally, vertically and diagonally (two directions). In the second step, the input pattern is searched from each side of its inner 30 x 30 area as shown in Fig. 4. The search proceeds until either a center line or a black pixel is reached, and its depth is measured. Each side of the 30 x 30 area is divided into six sections, and the measured depth values are accumulated for each section to extract 24 (4 sides x 6 sections) feature values. Discriminator 18 performs a first preliminary classification using the feature values obtained in the first step and corresponding feature values of reference patterns stored in read-only memory (ROM) 20. Any category having a distance Dl greater than a predetermined threshold value is discarded. The distance Dl is calculated as follows:

(Image Omitted)

wherefi: i-th feature value of a reference pattern gi: i-th feature value of an input pattern i : local contour direction In this calculation, the feature values are accumulated only for each local direction, and the 24 regions shown in Fig. 3 are disregarded. Discriminator 18 next performs a second classification using the feature values obtained in the second step. A distance D2 is calculated as follows:

(Image Omitted)

wherei: side number j: section number mf: average feature value of a reference...