Automatic Extraction of Product Regions from Images on C2C Online Market
Takuya Futagami, Noboru Hayasaka · 2019
A novel method that extracts only regions in products from product images used in consumer-to-consumer (C2C) online marketplaces or auctions is proposed. The proposed method automatically extracts product regions by using GrabCut based on fact that product regions tend to be found in the center part of images. Although GrabCut is known as an interactive segmentation technique, the proposed method automates procedure of extraction by generating an initial seed of GrabCut. To evaluate effectiveness of the proposed method, with 138 product images including 114 images obtained from a C2C online marketplace, the extraction accuracy of the proposed method was compared with that of the interactive GrabCut. The result shows that the extraction accuracy of the proposed method is 2.3 % lower than that of the interactive GrabCut. However, statistically significant difference in the extraction accuracy disappeared. In other words, the proposed method is effective in enabling automation for extraction without significantly degrading the extraction accuracy.