Mobile product recognition with efficient Bag-of-Phrase visual search

This paper presents a mobile product recognition system using bag-of-visual phrase (BoP). It aims to develop a mobile product recognition and recommendation system where a user can recognize a commercial product of interest by taking a picture of it using the mobile phone, and then search for the relevant information (e.g., price, nearby store, consumer recommendation, etc.). In the proposed BoP framework, second-order visual phrases candidates are first obtained from neighborhood visual words. Discriminative visual phrases are then determined, and images are indexed with a two-dimensional inverted index of visual phrases. Geometric verification (GV) is performed to further improve the accuracy of image matching. Experimental results show that the proposed method can achieve 90% recognition rate for a dataset consisting of 3882 reference images and 41 categories.