Character identification technique based on Gabor filter set

Abstract

A character recognizing technology based on Gabor filter group includes using Gabor filter group to extract the strokes of a character image in different directions, finding out optimal filter parameters, non-linear post-processing to the output of Gabor filter group for increasing the resistance of recognizing system to brightness, contrast change and interference of image, calculating the recognizing characteristics of the positive and negative values in said output, combining them, together to obtain a multi-dimension characteristic vector, and reducing its number of dimensions by linear analysis method to improve its recognizing performance. Its advantage is high recognition percentage.

Description

technical field [0001] The character recognition method based on the Gabor filter bank belongs to the field of character recognition (referred to as OCR). Background technique [0002] The traditional character recognition technology is based on the character image after binarization, so when it is applied to various low-quality images, such as low-resolution images, ID card images, car license plates, and natural scene images, due to the binary The character image quality after valueization is low, and the recognition performance is poor. Therefore, many techniques directly extract recognition features from grayscale character images. Generally, terrain features, gradient edge features, etc. are extracted by analyzing the gradient distribution of the gray image. These methods require the image brightness distribution to meet some specific rules, and the anti-interference ability is very poor; or use simple orthogonal transformation methods such as FFT and DCT. These trans...

AI Technical Summary

Problems solved by technology

Generally, terrain features, gradient edge features, etc. are extracted by analyzing the gradient distribution of the grayscale image. These methods require the image brightness distribution to meet some specific rules, and the anti-interference ability is very poor; or use simple orthogonal transformation methods such as FFT and DCT. These transformations reflect the global information of the image, but cannot extract the local stroke direction and structural information, so the ability to resist image brightness changes and character deformation is also very poor.

[0003] Because the Gabor filter has excellent time-frequency characteristics and the ability to effectively extract local structural features, it is also possible to use the Gabor filter to extract the recognition features of handwritten digital images, but this method has the following disadvantages: when designing the Gabor filter bank , according to the recognition rate to screen the parameters, the process is cumbersome, the amount of calculation is large, and the recognition performance is also poor; based on the binary image, the problem of applying the Gabor filter to the low-quality grayscale character image recognition is not considered, and the character details poor resolution

This method does not give full play to the excellent performance of the Gabor filter and cannot be applied to actual character recognition tasks

Images