Exact clothing retrieval approach based on deep neural network

摘要

In recent years, the trend of shopping for clothing on an onscreen character has been raising. Unfortunately, clothes that appear on screen could be difficult to track down, much less to purchase directly. It is a challenging task to find out exactly what the clothing in a TV show is from massive amounts of clothing items in online clothing retails due to the clearly noticeable visual discrepancies of the same clothing in video and in online shop photos. In order to tackle the problem, we adopt the deep neural network to learn recognizable clothing-specific feature for finding the exact clothing item in the video scenario among a large collection of online shop images. The experiments shows the clothing-specific feature is more effective in retrieval task and color filtering improve the retrieval performance.