A Comparative Study on CRFs for Fore- and Background Classification

Text contained in images and video frames provide an important clue for content based images and video indexing. However, in natural scene, text elements are corrupted by many types of noise, such as streaks, highlights, or cracks. These effects make the clean and automatic segmentation very difficult and can reduce the accuracy of further analysis such as optical character recognition. This paper uses different optimization methods for text segmentation from images with complex background based on conditional random field. Experimental results demonstrate the performance of the different methods. It enables the segmentation of text in complex situations more precisely comparing to the binarization of the image.