Research On Automatic Clothing Image Annotation Method

With the advent of the information age, a number of fashion clothing site clothing brand images surge, which have been many diverse styles and no clear classification, leading garment workers to pick out suitable to meet the design requirements, to bring their own creation inspired design spent more and more time and effort. How do clothing-related workers quickly and accurately retrieve their satisfaction images from the mass of clothing image library is a hot research field. Due to common understanding deviation between the image of clothing underlying visual features and high-level semantic concepts, artificial clothing label image has been difficult to meet the needs of users, in order to achieve better image retrieval clothing, clothing automatic image annotation increasingly occupy a pivotal role.Clothing automatic image annotation technology uses the trained classifier model to establish contact between underlying visual features and high-level semantic concepts, the clothing images which will be annotated achieve automatic image annotation based on the establishment of a good relationship model. This paper has been studied automatic image annotation clothing and related technologies on the basis of the classification model technology, mainly including visual features extraction and algorithm description both of color and texture features, after that the extracted visual features merge together as an integrated feature vector as the experimental data. Then through researching support vector machine theory and technology and on the basis of a large number of experiments, we propose a multi-core SVM clothing image automatic annotation method, combining with cross validation tuning parameters to achieve the clothing image classification based on SVM and labeling system which significantly improves the accuracy of clothing automatic image annotation.