Efficient Recognition of Objects by Cascading Approximate Nearest Neighbor Searchers

For object recognition based on nearest neighbor search of local descriptors such as SIFT, it is important to keep the nearest neighbor search efficient to deal with a huge number of descriptors. In this report we propose a new method of efficient recognition based on the observation that the level of accuracy of nearest neighbor search for correct recognition depends on images to be recognized. The proposed method is characterized by the mechanism that multiple recognizers with approximate nearest neighbor search are cascaded in the order of the level of approximation so as to improve the efficiency by adaptively controlling the level to be applied depending on images. From experimental results with 10,000 images, we have confirmed that the proposed method is capable of achieving a recognition rate of 98% in 1 ms / query, which is 1/10 of the recognition time without the cascade, and 1/40 of the recognition time with conventional approximate nearest neighbor search such as ANN and LSH. In addition, a recognition error rate of the proposed method has been suppressed to 0% by allowing a rejection rate of 8.6%. Experimental results with 100,000 images show high scalability of the proposed method.