TH105604B - A process for semi-automatic character image recognition. - Google Patents

Abstract

------February 21, 2020------(OCR) Page 1 of 1 Abstract of the Invention This invention presents a semi-automatic character image recognition method. This method recognizes characters where the user must input character data before use. The system then recognizes the character image by calculating the proximity of the input character image to the character initially specified by the user. If two or more characters are connected or are incomplete due to broken character strokes, it uses character characteristics such as character width to determine the character size and relies on the probability of character pairs to predict which adjacent character should be identified, thereby improving recognition accuracy. Furthermore, the proposed method can be applied to any language. ------------ DC60 This invention presents a semi-automatic character image recognition method. This method recognizes characters where the user must input character data before use. The system then recognizes the character image by calculating the proximity of the input character image to the character initially specified by the user. If two or more characters are connected or are incomplete due to broken character strokes, it uses character characteristics such as character width to determine the character size and relies on the probability of character pairs to predict which adjacent character should be identified, thereby improving recognition accuracy. In addition, the proposed method can be applied to any language. This method uses unique characteristics of characters, such as character width, to determine character size and relies on the probability of character pairs to predict which adjacent characters should be, thereby improving recognition accuracy. Furthermore, the proposed method can be applied to any language.

Claims (1)

------February 21, 2020------(OCR) Page 1 of 2. Claim 1. A semi-automatic character image recognition method to convert character images into character strings in a system consisting of a bigram (2-grams) database and a font template database. The method consists of the following steps: i) Receiving image data containing a number of character images; ii) Preparing the image data by dividing the character image into character strings, each string containing at least one character, and separated by spaces; iii) Checking the size of the character string. If the string size is greater than the specified size, it means that the string contains more than one character. The string is then divided into a first character image segment and a second character image segment; iv) Comparing the first character image segment of the character string with the font template in the font template database and selecting the closest character to be the first character; v) Searching for the second character pair from the bigram (2-grams) database based on the selected first character, selecting the character pair with the closest value. The highest probability in the first bigram database is used to compare the character image from the corresponding character style database with the second character image, and vi) repeat step (v) using the next highest probability character pairs from the bigram database until a character image that matches the character in the character style database or the character in the character image database that gives the highest similarity is found. 2. The method according to claim 1, which also includes a step to calculate the average character spacing (space) and the average character width. 3. The method according to claim 1, where the image data acquisition step is performed by scanning the second page of two pages. 4. The method according to claim 1, where the image data preparation step also includes performing horizontal projection to divide the character image into a number of rows. 5. The method according to claim 1, where the calculation of the similarity of the character image with The font style (template) uses the following equation: MAX Mi = (symbol)(Tij-Cij), where Mi represents the proximity value at row i, Tij represents the color value of the template image at row i and column j, and Cij represents the color value of the character image to be recognized at row i and column j.