Text extraction from assorted images using morphological-region, texture and multiscale techniques - A comparative study

Text embedded in image provides high-level semantic information for automatic annotation, indexing and retrieval. Extraction of text is difficult due to the complex background in image, font size, shape, orientation, as well as color variance, and these make it a challenging task. In recent years many methods have been proposed for text extraction in different images. In this paper the comparative evaluation analysis of various text extraction techniques is presented. It has been shown that the Morphological - Region combined algorithm is computationally more expensive and performs better than the texture based and multi-scale edge based text extraction methods under almost all scenarios.