New Approach to Low Contrast Image Segmentation

Although some breakthrough has been made on image segmentation by using level set based curve propagation techniques, however, these approaches usually are unable to segment exactly the images with low-contrast boundaries or edges. In particular in medical image processing, low-contrast images sometimes are unavoidable due to capturing devices, noise, or partial volume effects. Motivated from the background model on segmentation of moving image sequences, a new approach, in this paper, is proposes to remedy this problem. To reduce the effect of different contrast or intensities, a weight function is defined and applied to each pixel of the image, which trades the effects of geometric and photometric. The weight only relies on the relationship between a point and its neighborhood ones. By means of the help of the weight function, image segmentation may be performed on a weight map so that the side effects resulted from the nonhomogeneity of intensities in the regions are greatly reduced. In contrast with existed approaches, the proposed approach is fast and with very low computational cost. Moreover a high flexibility also is obtained by applying different diffusion functions on computation of the weight. The proposed algorithm has been validated with some numerical results.