Text Segmentation And Recognition In Natural Scene

Natural scene text images contain rich information, which is important to scene understand, scene analysis and image retrieval. Natural scene text analysis has always been a challenge due to its complex background, random distributed texts and different screen angles.This paper focuses on the text segmentation and text recognition, and the main research is done as follows:1. text segmentationThe aim is to segment the background and the foreground of the text area obtained after the text location. Several methods have been introduced and a comparison has been made to show the performance of different methods. An improved Bradley method has been proposed to deal with the low contrast and uneven lighting images. As for complex background images, the color information has been considered and a method using color and text analysis is employed.2. character recognitionTwo researches have been done on the text recognition. The first research is based on the idea of coarse classification and fine classification. In this paper, a side profile feature is been used to get a coarse result and the direction pixels feature is obtained to get the final result.In the second experiment a Gabor transform is used to get the feature of characters and the Angle cosine vector is employed as the classifier. The experiment results also show the influence of different parameters of Gabor filter, different classifiers and different grids on the recognition performance. And the conclusion that the Gabor transform is effective can be safely drawn.