Research On Leather Classification Based On Color Features

Leather is one of important supplies in people daily life, is also an important export commodity. China is a big country not only in leather production, but also in leather consumption. Leather processing industry as one of the pillar industry of China’s light industry, has experienced the initial advantage of low-cost labor force. The initial mode of production can not be maintained under the pressure of rising labor prices of production. Research in machine vision technology to replace labor, increase productivity and reduce costs, has great significance for upgrading leather processing industry.The thesis mainly focuses on leather image classification technology based on color features, putting forward some measures on color feature extraction, similarity calculation and classifier, finally achieving the classification system of leather color. The main work of this paper is as follows.1. For the inherent characteristics of leather image, the paper presents an improved main color clustering algorithm. Firstly, according to the characteristics of the human eye color perception, converting the RGB color space to the improved HIS color space, choosing the pivot point, clustering the main color. Finally performing the KNN nearest neighbor classification by using the weighted distance as the similarity calculation method. Experiments show that the algorithm can effectively distinguish the leather image in color.2. Aiming at the existence of natural texture in leather image, affects the accuracy of color classification of leather image. This paper proposed a relative to the total variation model to remove the texture of leather image. Then use the average color components( L* a,* b,*) to represent color features based on (?) color space’s strong ability to distinguish color. Finally, use the support Vector Machine to classify the leather image. Experiment show the algorithm is more accurate than improved main color clustering, but it’s computational complexity is too large, relatively weaker than improved main color clustering in real time.3. In this paper, we design and implement a leather color image classification system in hardware and software. In this system, we realize the algorithms mentioned above.