A new approach for segmentation of image and text in natural and commercial color documents

This paper presents an efficient method for segmenting text and non text parts of natural real life images and colored document images using M-band wavelet packet frames. Various combinations of band pass channels of M-band wavelet packet frames represent the image at different scale and orientations in the frequency planes of YCbCr components of color images. The scale space feature vector comprises of the local energy around each pixel at different scales and segmentation is achieved using fuzzy C-means clustering. No information regarding font size, scaling representation, type of layout etc. of the images are considered in our algorithm.