Enhanced Quantization Based Bag-of-Features Model And Its Applications

Bag-of-Words (BoW) model has been widely used for various problems in the fields of multimedia and computer vision. The key idea of BoW model is to quantize the local descriptors of images to visual words, so as to represent each image by a his-togram of the visual words. Usually, a clustering algorithm is first used to group the local descriptors into several clusters. In the quantization, local descriptors are then mapped to one or more visual words by referring the cluster centers. In this paper, we argue that simply using the cluster centers to represent the visual words may lead to unreasonable mapping. It is more reasonable to further capture the cluster size to represent the visual words in the quantization step. To this end, we introduce a simple yet effective coding algorithm called Effective Radius Coding (ERC), in which the effective radius of the cluster is introduced together with the cluster center to repre-sent a visual word. To effectively estimate the effective radius, multiple anchor points supporting a cluster are computed to capture the cluster size. Moreover, we extend the traditional k-means method using effective radius to enhance the visual words generation process. Experimental results demonstrate that our proposed algorithm improves the Bag-of-Words model and outperforms the state-of-the-art techniques.