Fusing multiple features and spatial information for image classification via codebook ensemble

Abstract

The construction of a codebook is an important step which is usually done by cluster analysis. However, clustering is a process that retains regions of high density in a distribution and it follows that the resulting codebook need not have discriminate properties. This paper presents a discriminative spatial codebook ensemble learning approach for image classification with three key innovations: 1) images are first divided into sub-regions according to a spatial pyramid, and then initial big member spatial codebooks are constructed by grouping features of sub-regions into a number of clusters, one member spatial codebook for one sub-region; 2) the discriminative member spatial codebook is formed by selecting the visual words with higher probability of occurring in the images. Then the features of each sub-region are coded by LLC based on its corresponding member codebook; 3) combining SIFT and KDES-G features to describe images is also proposed by generating a joint vector as a new feature vector. The experimental results on the Caltech101 and 15 scenes datasets have shown that the proposed method has better performance and robustness compared with some state-of-the-art works.