Sparse Representation Classification for Image Text Detection

Abstract

Text detection in images is important for the retrieval of text information from digital graph, video databases and web sites. In this paper, a text detection method based on sparse representation classification with discrimination dictionaries is presented, which can detect text with different sizes, fonts and colors. The propose method detects edge information using Sobel operator and a sliding window scans the edges into patches to facilitate sparse representation process. Then the roughly text area is detected by sparse representation classification based on discrimination dictionaries. Finally, a projection profile analysis is used to refine the detected text areas. The detection performance of our approach is tested using a set of video frames taken from MPEG-7 video test set.