urrent OCR (Optical Character Reader) systems cannot accurately recognize low-resolution printed kanji characters because low-resolution character images involve many jags, blurred or thinned areas after binarization is performed. In this paper, a restoration method is proposed in order to obtain a high-quality binary character pattern from a low-resolution character image. First, the character image is expanded and blurred before being converted into the binary character pattern. Next, the binary character pattern is modified by a system using ridge and ravine points. The feature vectors of the modified character pattern that includes the spatial structure information are extracted by a phrase feature distribution method and they are fed into a classifier. It was verified by experiments using the database that the proposed method improved recognition accuracy.
展开▼
