Pin-Syuan Huang, Jing-Yi Tsai, Yu-Fang Wang and Chun-Yi Tsai, National Taitung University, Taiwan, R.O.C.

The objective of this paper is to propose a combinatorial encoding method based on VLAD to facilitate the promotion of accuracy for large scale image retrieval. Unlike using a single feature in VLAD, the proposed method applies multiple heterogeneous types of features, such as SIFT, SURF, DAISY, and HOG, to form an integrated encoding vector for an image representation. The experimental results show that combining complementary types of features and increasing codebook size yield high precision for retrieval.

VLAD, SIFT, SURF, DAISY, HOG