LAUSR

ABSTRACT Consumer-to-consumer (C2C) online market places have become popular. Several C2C online market places adopt product recognition from uploaded images representing the current state of the products to aid in the entering of product information for creating listing pages. To improve recognition accuracy, it is important for extracting product regions from product images as a pre-processing for recognition. Given these circumstances, this study proposes a method of extracting product regions from images used in C2C online market places. We analyzed product images for effective product extraction and developed the proposed method using the region-growing algorithm and GrabCut segmentation algorithm based on these analysis results. To generate initial seeds for GrabCut, the proposed method specifies image-border areas as background areas based on the analysis results and applies the region-growing algorithm to the specified background areas. To evaluate the effectiveness of the proposed method, we compared its extraction accuracy and computational time with those of a conventional method using 412 product images, including 341 actual images. The proposed method was effective in both extraction accuracy (20.3% improvement rate) and computational time (76.7% reduction) compared with the conventional method. Compared with the conventional method, the proposed method increased the extraction accuracy for all the product categories from sellers. Therefore, the effectiveness of the proposed method can be observed for several product images. Furthermore, we confirmed that each process of the proposed method is necessary for improving the extraction accuracy.