Authors : Zahraa Faisal and Nidhal K. El Abbadi
Abstract: Skin cancer lesioning one of the most skin lesions which cause death. Lesion segmentation is a very important step prior to detecting and classifying the skin cancer. In this study, we introduce a new method to extract the lesion from the surrounding of healthy skin. The proposed method starts with image preprocessing included image de-noising and removing the unwanted objects such as thin hair and air bubble by using the median filter, followed with edge detection using the Markov and Laplace filter. The current algorithm converts the color image to YUV color space and select the U channel for processing. Thick hair is removed from U channel by combining both morphological operation and median filter. Mathematical morphology such as close used to join narrow breaks regions in an object, fill the small holes and remove small objects. The final step is to find threshold based on Otsu�s thresholding to separate the image to two regions one for lesion and the other for skin. The result image is binary image or can be color lesion with black background. The accuracy of the suggested method reaches up to 98%. The algorithm is tested with segmented images by an expert and give very promised results in many cases gives better results. Also hamming distance is imeasured and it was better value compared with other algorithms.
Zahraa Faisal and Nidhal K. El Abbadi, 2017. New Segmentation Method for Skin Cancer Lesions. Journal of Engineering and Applied Sciences, 12: 5598-5602.