Research On Image Classification Algorithm Based On Emotion

With the massive growth of the internet images, as well as the rapid development of Human-computer interaction systems. How to effectively classify and organize of these images, so that people can find the images that they really want in the vast image resource, has increasingly attracted attention. In recent years, image emotional semantic classification has become an active research topic in computer vision. As a high-level semantic classification, it is of great significance for the correct interpreting of image content and the effective organization, classification and management of image. How to effectively compensate for the "semantic gap" between the image of low-level visual features and high-level semantic features of the image and build the human perception of emotional mapping mechanism is the key to achieve the emotional classification of image. In this paper, we combine the knowledge of visual cognition, psychology and pattern recognition into research on emotion of emotional classification of images.Based on the visual word bag model (Bag-of-words), this paper presents a salient regions weighted emotion classification algorithm. First, the algorithm uses CSIFT (Colored scale invariant feature transform) features to generate emotional visual words, Compared with the previous SIFT features generated visual words, CSIFT joins the color information which has a strong semantic discrimination. Second, we extract the image saliency map information to determine the different salient degree of emotional visual words, and thus giving different weights to emotional visual words. This method can enhance the weight of those emotional visual words that have more impact on human emotion.At the same time, taking into account the color as an important features, we extract global HSV color histogram, and then fuse color histogram with the previous weighted emotional visual vector. Finally, support vector machine classifier is used to complete the emotional classification. Experimental results demonstrate the effectiveness and superiority of our approach.