Locality-Constrained Linear Coding Based on Principal Components of Visual Vocabulary for Visual Object Categorization

Through the linear correlation analysis between the local feature and its K-nearest-neighbor visual words and significance testing of locality-constrained linear coding, this paper finds that the fundamental reason for causing nonsignificance of the weight coefficient is the multicollinearity of K-nearest-neighbor visual words in Locality-constrained Linear Coding (LLC) scheme. Locality-constrained principal component linear coding can solve the multicollinearity and improves the classification accuracy, but it increases the time overhead of the coding. This paper presents an improved scheme called Locality-constrained linear coding based on the principal components of visual vocabulary. To determine the principal components of K-nearest-neighbor visual words of each local feature is simplified to only determine the principal components of visual vocabulary. Experiments have been conducted for comparing and evaluating the proposed method utilizing the Caltech-4 dataset. Experimental results show that locality-constrained linear coding based on the principal components of visual vocabulary reduces the time overhead and the same time it retains the advantages of Locality-constrained principal component linear coding.