Character extraction and recognition for low-resolution color images using dominant-color-based-line-segment method

摘要

A new extraction and recognition method for low-resolution characters in complex color images has been proposed. This method generates contributivity images with region segmentation based on extracted dominant colors and recognizes characters in the multi-scale contributivity images. The contributivity images are generated using Dominant-Golor-based-Line-Segment Method which decides pixel values based on contributions of dominant colors to the pixel colors by calculating distances between the pixel colors and line-segments through pairs of dominant colors. Experiments using web images show that the proposed method has increased extraction rate from 77% to 97% and recognition rate from 62% to 85% as compared with a traditional method using k-means clustering and binary character recognition.