Study of Copyright Protection for Merchandise Pictures in E-Commerce

For solving the picture misappropriation problem in e-commerce, this paper proposes a content-based copyright detection method for e-commerce website pictures. First, a picture is smoothed by bilateral filter in order to effectively preserve the key points in the edges and low-contrast regions, which ensure the integrity of its key points. Second, the picture´s texture feature is extracted by using Gabor filter, and the shape feature is captured by Hu moment. Third, feature vectors are assigned weights and normalized in order to reduce the effects of variations in their dimensions and components. Finally copyright-infringing pictures are detected by calculating the Euclidean distance and we use fuzzy heuristics to measure similarity between the query and the database images. The test dataset is crawled from the largest e-commerce website (Taobao.com). The copyright-infringing pictures can be detected by the method and the average accuracy reaches 91%, achieved the desired effect.