Research On Clothes Detection And Recognition

With the development of digital televisions, the applications of various interactivetelevisions are becoming more and more popular. To provide better user experience andcommercial value, the technology of image recognition is of great importance to thoseapplications. Currently, compared with face recognition, clothes recognition has beenextensively studied while deserves in-depth investigation. In this thesis, a method ofadaptive detection and recognition of clothes is proposed to identify the clothes in colorimages.Since there are little common characters for various clothes, a novel clothesdetection algorithm which combines the face detection and information of clothes edgesis proposed. In this method, the possible region of clothes is firstly determined byemploying the AdaBoost face detection algorithm. Subsequently, according to theclothes edge information, an iterative algorithm is designed to find the rectangles whichcontain most information of cloth edges, and then the accurate location of clothes can beidentified. The experiment results indicate that this algorithm can extract the region ofcloth effectively.Furthermore, considering the highly dimensions of clothes images, a scheme ofclothes recognition method based on Principal Component Analysis (PCA) is developed.With this method, the dimension of clothes images can be effectively reduced. Then thecolor histogram based method for clothes recognition, which utilizes the invariableproperty of the color, is presented. Finally, the method which combines the PCA andcolor histogram analyses is proposed. The experiment results demonstrate that thismethod can achieve advanced performance for clothes recognition.