Image retrieval based on the feature of VLAD and sparse representation

Abstract

In order to achieve the goal of fast and accurate image retrieval,an image retrieval method combining VLAD （vector of locally aggregated descriptor）feature and sparse representation was proposed. Firstly, according to the characteristics of rich structure details and obvious differences for local visual features in image, the local rotation invariant SURF feature of the image was extracted,and the local VLAD feature of the image with rotation invariance was constructed by the local aggregation descriptor method. Then, the VLAD feature was combined with the sparse representation（SR）to design the similarity retrieval metric based on SR,thus the retrieval of the image could be realized. The experimental results show that,proposed method outperforms the compared methods in terms of precision,average normalize modified retrieval rank（ANMRR）and other indicators,and it also has higher computational efficiency.

Keywords

• image retrieval;sparse representation;vector of locally aggregated descriptor