CNN-SIFT Consecutive Searching and Matching for Wine Label Retrieval

摘要

Wine label retrieval is key to automatic wine brand search through the web or mobile phone in our daily life. In comparison with the general image retrieval tasks, it is a rather challenging problem with a huge number of unbalanced wine brand images. In this paper, we propose a CNN-SIFT Consecutive Searching and Matching (CSCSM) framework for wine label retrieval. In particular, a CNN is trained to recognize the main-brand (manufacturer) for narrowing the searching range, while the SIFT descriptor is improved by adopting the RANSAC and TF-IDF mechanisms to match the final sub-brand (item attribute under the manufacture). The experiments are conducted on a dataset containing approximately 548k images of wine labels with 17, 328 main-brands and 260, 579 sub-brands. It is demonstrated by the experimental results that our proposed CSCSM method can solve the wine label retrieval problem effectively and efficiently and outperform the competitive methods.