An image segmentation scheme based on color clustering and wavelet transform

A novel multi-scale segmentation scheme based on color clustering and wavelet transform is proposed in this paper. Firstly, a color clustering method is performed to segment images at a coarse scale. Secondly, the high frequency wavelet coefficients and their statistics are used to perform classification of image blocks at fine scales. Post-processing will refine the boundaries of foreground and eventually extract the objects. This novel method is applicable to ordinary images and has achieved high accuracy when tested on images with various background. Experiments show that the new algorithm provides high accuracy and the computer cost is quite short.