Text Segmentation From Images With Textured and Colored Background

In this paper, a method to segment the text regions from color images with textured background is proposed. The method is based on finding the text edges using decomposed color information content of the subimage coefficients obtained from the discrete wavelet transformed input images. Detected edges are then combined in order to form the exact location of the characters. In the next stage, the regions that are not acceptable as the text regions (due to some general structural rules) are removed to improve the total performance. The final outputs of the algorithm are the characters on a smooth background. The experimental results show that the proposed method is robust to size, font, language, color and direction changes of the text regions.

Keywords:

Text extraction , text segmentation , wavelet transform , image documents , PCA