An Approach to Edge Detection in Images of Skin Lesions by Chan-Vese Model

• Author(s): Roberta B. Oliveira ; João Manuel R. S. Tavares ; Norian Marranghello ; Aledir S. Pereira
• Date: 2013
• Persistent ID: http://hdl.handle.net/10216/65338
• Origin: Repositório Aberto da Universidade do Porto
• Subject(s): Ciências Tecnológicas
• Document Type: Book
• Language: English

Description

Nowadays there is a great interest in the application of computational systems for the analysis of skin lesions. These systems allow the dermatologist to prevent the development of malignant lesions. The development of the systems has occurred due to the increase of skin cancer cases. In the characterization of skin lesions it is necessary to segment the images accurately. Thus the features and edges information of the lesion can be extracted and used by a classifier or by a dermatologist for a better classification. When images are acquired in a non-systematic and non-controlled way there may be a segmentation problem. In this case the skin lesion of images can have different sizes and various type of noises, such as the hair. These factors can affect the detection of the lesion edges and complicate its characterization. One solution would be to apply a smoothing filter to reduce noise before the segmentation step. Segmentation techniques adapted to each type of image can be used to solve the problem of diversified images, such as images with different sizes lesions, reflexions and light intensities. In this paper is proposed a computational method to assist the dermatologists in the diagnosis of skin lesions by digital images. It was used the anisotropic diffusion technique for the preprocessing of the images in order to remove the noises. The Chan-Vese model was used to segment the lesions. The next step consists of the application of morphological filters to eliminate outside and inside noises from the object, that remained in the segmented images, and also to smooth their edges. This approach allowed to minimize noise problems and edge detection to different cases of skin lesions images, such as melanoma, melanocytic nevi and seborrheic keratosis. The segmentation achieved 94.36% of accuracy for the three types of skin lesions.