A Statistical Approach to Improve the Human Dermoscopic Image Classification by Using Various Non Linear Edge Preserving Image Enhancement Methods

Abstract

Skin cancer is considered one of the most dangerous types of cancer, with a sharp increase in mortality due to a lack of knowledge of symptoms and prevention. The number of patients suffering from skin cancer is increasing worldwide. Timely and precise identification of skin tumours is important for reducing mortality rates. An expert dermatologist is required to handle cases of skin cancer using dermoscopy images. Improper diagnosis can cause fatality for the patient if it is not detected accurately. In order to stop cancer from spreading, early identification at an early stage is required. Skin cancer is mainly classified into two main types: Benign and Malignant. Some of the classes fall under the category of benign, while the rest are malignant, causing severe issues if not diagnosed at an early stage. Automatic classification of skin lesions is always a challenging task due to the different shape and size of tumours, low contrast, light reflections from the skin surface, etc. The survival rate of people affected by skin tumours can be increased if detected early, so a method for the classification of skin tumor images is proposed. In the first stage, the input image is decomposed into a base and detailed layer, and bilinear interpolation is applied to both layers. Then the detailed layer is amplified with enhancement ratios of 1.5 and 2. Then both layers are merged, and then the bilateral filter and WLS filter are applied separately. The enhanced images are assessed by metrics: MSE, PSNR, SSIM, and Entropy. In the second stage, enhanced images are segmented using the k-means clustering method, and features are extracted using the statistical feature technique Gray Level Co-occurrence Matrix (GLCM) method from the segmented images. After obtaining the features, the enhanced images are used for classification by training the tumor images from the dataset, and the enhanced images are used for testing. With the Bilateral filter as a pre-processing stage and the SVM and CNN classifiers, we are achieving classific