| Popis: |
According to an estimate made by the National Cancer Institute (INCA) in 2012, also valid for the year 2013, the skin cancer appears as one of the most cancer types common in Brazil. The high level of predominance of the skin cancer case has motivated the search and the development of computational methods to assist dermatologists in the diagnosis of skin lesions. The main goal of such methods is concerned to the detection of benign skin le- sions to prevent their development, or diagnose malignant lesions at early stages so that they undergo appropriate treatment plans with higher chances of cure. The objective of this paper is to present a computational method for extracting edges of skin lesions from photographic images in order to facilitate the extraction of its main features used for classification. This paper presents a method for the extraction of contours of skin lesions, such as nevi, seborrheic keratosis and melanoma, from images, which uses the technique of anisotropic diffusion to smooth the input images and the active contour model without edges, known as Chan-Vese model, to segment the smoothed image. The application of the anisotropic diffu- sion filter removes selectively the noise present in the input image. The Chan-Vese model is based on the Mumford-Shah region growth technique, common used in image segmentation tasks, and the Level Set Active Contour model, which allows topological changes of the curves applied on the input images to segment them. Then, a morphological filter is applied on the segmented images in order to eliminate holes in the skin lesion regions and also to smooth their edges. Experimental tests have been accomplished to compare the segmentation results obtained by the traditional thresholding method, by the combination of an anisotropic diffusion model and the Chan-Vese model and by the proposed method using grayscale der- matologic images. This comparison has been revealed that the method proposed is effective to detect skin lesions and extract their contours in dermatologic images. |
