Color recognition of clothes based on k-means and mean shift

According to the problem of color recognition of clothes for image searching on web, a kind of clustering main color detection with one step of k-means and one step of mean shift is adopted and the color of clothes can be detected accurately. It uses the k-means to extract the background and foreground of a picture, which contains the information of a person dressed up with a color cloth in a complex background. The first step aims to recognize the position of the cloth content and lows the influence of the background. After the first step of k-means, we reconstruct the image as the clustering data for the second step of mean shift. The recognition effect achieves above 85% by calculating the distance of the clustering center of our color model of pixel, comparing to the database with the accurately color map of our clothes´ images. The experiment of our approach shows that the two-steps k-means and mean shift of color recognition is improved. We developed a color recognition system of clothes for the image searching using that approach and the recognition performance is optimized.