Principal object detection towards product image search
Xiao Ying Wu, Ling-Ling Liang, Wenjian Wang, Qiang Peng · 2012
Online shopping is an attractive, convenient, and efficient shopping way for billions of web users. A disappointing fact is that it is usually difficult for users to find the products fitting their needs purely based on text search. Content-based product image search becomes a promising way to solve this problem. However, the presence of natural backgrounds and fashion models significantly affect the feature matching, which makes product image search a challenging task. To clean the background and minimize the influence of noises, in this paper, a graph-based principal object detection algorithm is proposed to extract the product items while removing backgrounds and noises. A product image retrieval system is then constructed to verify the effectiveness of the proposed approach. Experiments on a large scale dataset with 1.36 million product images crawled from Taobao demonstrate the proposed approach significantly improves the retrieval performance.