Publications

Cancer describes a type of malady distinguished by uncontrolled growth and division of abnormal cells. Skin cancer is one the most commonly diagnosed cancer. Image-based computer aided diagnosis (CAD) systems have significant potential for screening and can help discover cancer in its earlier stages. The bottleneck of the CAD system is the skin image segmentation. Image thresholding techniques are the most used for image segmentation. Statistical approaches are widely used in image thresholding. Due to the simplicity of its mathematical formula, different skin image thresholding techniques used Gaussian distribution as a model of data image. However, that distribution has a limitation when the histogram modes of the image has non-symmetric shapes. Gamma distribution has symmetric and non-symmetric shapes and has been used to improve the skin image thresholding. That technique assumed that the shape of each mode in the histogram is constant. However, the shape of each mode in the skin image histogram can be vary inside the image itself. In this paper, our contribution is to use different parameter shape for each mode in order to improve the quality of skin cancer image segmentation. Experimental results showed that the improved technique has better results than existing techniques using performance measures.