Abstract: Due to the increased demand for the diversification of clothing products, clothes
classification is very necessary, no matter for operators or consumers. Existing methods usually solve
the problem based on the whole clothes, paying little attention to the detail features on the clothes.
Therefore, identifying and classifying the detail features of clothes are emphasized in this paper such
as the type of collar, the length of sleeves and the trousers or the dresses. Based on the contour
extraction, the paper proposes the voting strategy for the results gained from multi-scale HOG
features. Also the geometric features are used based on corner detection, solving problems like collar
position uncertainty, neckline shape interference caused by surrounding patterns and so on. Then,
SVM classifier is used to get the final results. After that, some advice is also provided on costume
matching using multiple coefficient matrixes of features matching. Experiments show that our method
is effective for classifying some mentioned detail features. Also, it shows some practical value for the
automatic matching recommendation.

Key words: detail features of clothes, contour extraction, HOG features, geometric features, costume
matching