Interactive clothing retrieval system

We present an interactive clothing retrieval system, which supports query by a real-world image with target clothing and returns real-world images with similar clothing. The novel clothing shape feature is proposed to describe the shape of clothing in human-oriented coordinate system. We also propose a supervised method for learning a weighting matrix to minimize the intra-class distance while maximize the inter-class distance. The clothing retrieval results are reported quantitatively and the experiment results show that our system supports an accuracy of 61% over the dataset consist of practical various clothing types.