Nonuniform Illumination Image Segmentation Based on Improved Homomorphic Filtering and Class Uncertainty Theory

Because of the impact on image quality, it is difficult to segment fuzzy images under uneven illumination. This paper creatively combines improved homomorphic filtering method based on illumination reflection model, gamma transformation, class uncertainty and region stability together to segment nonuniform illumination images. Firstly, the illumination impact of the image would be restrained by improved homomorphic filtering, so that the quality of the image could be improved. Secondly, gamma transformation is applied to enhance the contrast of the image after the first step. Then, class uncertainty and region stability are used to measure numerical information and spatial information respectively. Finally, an energy function is structured combining class uncertainty and region stability to find the optimal threshold. The experimental results indicate that the introduced method could make better performance in some application scenarios than some other traditional methods and the method MUSEM we cited.