An Enhanced Approach for Skin Lesion Smoothening and Segmentation from Dermoscopic Images
Any abnormal growth in the skin is called a skin lesion. It could either be primary or secondary. Skin cancers such as melanoma and carcinoma fall under the category of precancerous lesions and can be treated if identified at the earlier stages. Computer aided diagnosis tools are used for skin lesion image segmentation and classification which helps to improve the process of identification of melanoma. The lesion edge segmentation is important to detect the infection precisely in dermoscopic images and diagnosis of diverse types of skin lesion. In this work, an enhanced approach for removal of noise followed by segmentation of skin lesions is proposed. In the enhanced approach, the dermoscopic image is pre-processed by applying bivariate shrinkage on the high frequency components obtained using dual tree complex wavelet transform. The obtained denoised image is then subjected to Fuzzy C means clustering for segmenting the desired region of interest. The dataset used is provided by the International Skin Imaging Collaboration (ISIC 2017) for public analysis and experimentation. The goal of segmentation is to find out the precise skin lesion area. In order to achieve good segmentation result, the image is first subjected to the maximum extent of noise removal. The denoised image thus obtained results in a better segmentation and a more convincing accuracy when compared against the ground truths. The experimental results showed that the proposed enhanced approach consistently produces the higher PSNR, Structure Similarity Index and Correlation Coefficient for Bio medical dermoscopic skin lesion images.
Keywords: Skin lesion; melanoma; wavelet transform; shift variance; divergence; fuzzy clustering; morphology; erosion; dilation; noise models