Skin Segmentation based on Improved Thresholding Method

Segmentation aims at partitioning area in the image based on color, shape or textures. It is useful in many computer vision applications such as medical image analysis, object detection and recognition, and content-based image retrieval (CBIR). Currently, CBIR has been applied in an online clothing purchasing system. Here, the goal of the segmentation process is to segment the clothing parts from the background and the model skin. Practically, we want to detect the skin color and segment it from the image, and left the apparel parts in the image for matching processes. We propose an improved threshold-based segmentation technique by adding a new constraint to the previous method to limit the skin tone selection. The technique improved the segmentation results by correctly segment the apparels with the shade of purple, pale pink, and pinkish-yellow. Simulations on total 100 garment images showed that the proposed method increased the segmentation rate by 27%.