Abstract: A novel text segmentation approach based on unsupervised clustering using color and stroke features was presented. Firstly, the possible text and background color is estimated on the enhanced text line image by color reduction and histogram calculation. Then color and stroke features were extracted and text and background pixels are determined by K-mean clustering algorithm. At last, segmentation result is optimized by post-processing. The performance of our approach is demonstrated by experimental results for a set of images with Chinese and English text

Key words: Text segmentation, Stroke feature, Unsupervised classification, Text color