A correlation-based bag of visual words for image classification

Image classification is a method that distinguishes the different categories of targets based on the different features of image. The current problem usually is that the feature modeling of target has a great influence on recognition robustness. In order to solve this problem, a correlation-based method is presented to optimize the bag-of-visual-word (BOVW) model by reducing the dictionary size. The features with strong relevance to categories are preserved to establish a visual dictionary. The modest visual dictionary is trained by support vector machine (SVM) classifier and its properties are analyzed. Finally, the effectiveness of the method proposed in this paper is validated through experiments. The experimental results demonstrate that the presented idea not only improves the robustness and accuracy of image classification, but also works well on practical problem.