Super-pixel segmentation based skin texture pattern recognition /

Abstract

Super-pixel segmentation is widely used nowadays in image processing to enhance segmentation accuracy. A new detection model is proposed for skin texture pattern recognition of Leopard, Cheetah, and Jaguar. In this model, a combination of Histogram of Gradient (HOG) and superpixel segmentation is used for extracting the features and the segmentation task of the target animal. This method does not require several superpixels to be created in advance, whereas it can automatically partition the image to its content into a suitable number of superpixels without any over or under segmentation. Then, the obtained features are fed into a Support Vector Machine (SVM) classifier to classify the skin texture pattern of Leopard, Cheetah, and Jaguar. The validation is performed which shows that the classifier achieves an accuracy of 96.67 %.