In the filed of character recognition, researchers have achieved great success in the recent years. The current optical character recognition (OCR) system only can handle the texts which are separated from the background and transformed to a binary image. When facing with complex background, low-quality image, they usually achieve poor performance. There exist two challenging problems, which are bottlenecks for enhancing rate of the character recognition. One is character recognition with complex background. It is necessary that the automatic system of text detection detect and extract text from the image which can be recognized by OCR. The other is degraded character recognition in low-quality. It is very difficult for character recognition that blurring, conglutination, broken stroke, low resolution exist in image. It is necessary to find a good way to resolve above two problems. Aiming at this goal, this dissertation involves with the following aspects: (1) By the direction of Metasynthesis theory, this dissertation proposes a method of text detection in image based on integration of multi-features, which integrates multi-features according to text multi-characteristic, and designs an automatic system of text detection in image. (2) In order to resolve the character recognition in low-quality image, this dissertation proposes a method of resolution evaluation of character image, based on gray distribution feature, and builds up the system of resolution evaluation of character image. Experiment result demonstrates the proposed approach can effectively evaluate the resolution quality of character image. (3) This dissertation applies the resolution information of character image on the degraded character recognition with multi-reolution. By integration technique of pattern recognition, the proposed method complements the low recognition rate and bad stability of single classifier to build up integrated net of multi-classifiers. The dissertation proposes a method of adaptive character recognition for degraded character recognition with multi-resolution, and designs the system of degraded character recognition with multi-resolution.
修改评论