A novel method for text and non-text segmentation in document images

Segmentation of the contents of document images into text and non-text regions is an essential pre-processing step for applications such as document analysis and classification, as well as OCR. This paper presents a novel technique to segment the document image into text and non-text regions using a combination of Wavelet-based Gray Level Co-Occurrence Matrix (GLCM) features and K-means clustering. A comparison between the performances of different wavelets in document image segmentation is also performed and tabulated. The technique was tested on a number of scanned article images from the MediaTeam Document Database and results show a marked improvement over the already existing method based on GLCM features.