Research On Visual-Perception Oriented Image Emotional Classification

Visual information is an important source of external information for human beings and an information form that is most easily accepted and understood by people.As an important form of visual information,images have been widely concerned and used by people.There is a big amount of images in the Internet.How to understand and deal with such a large-scale image data is an urgent problem to solve.The existing studies of the image understanding remain focused on the description of the image content,which belongs to the cognitive level understanding.However,with the exploding of images on the Internet,especially the widespread use of online social networks,users like to share a lot of information about their own feelings and experience.This makes the understanding of image content can no longer satisfy users.Therefore,understanding the emotional information in images has become more and more important and it is of great practical application value.This paper mainly focuses on the affective classification of images.Combining with the research of visual cognition theory and considering the fusion of multiple image features,we adopt the SVM method of machine learning to address the sentiment classification of images.The main work of this paper includes the following three parts:Firstly,optimizing the algorithm of color histogram with the Itti visual attention model to make the description of color information in images more consist with people's understanding and feeling.Secondly,extracting variety types of visual features that can express the emotional information of images,and training a SVM classifier to conduct an empirical study of the emotional classification of images in IAPS and Flickr data sets.Thirdly,applying the algorithm proposed in this paper to the application of movie recommendation.According to the sentiment analysis results of movie posters and synopsis to recommend the movie that meets the emotional needs of users.This paper optimized the algorithm of color histogram and proposed a complete architecture of image emotional classification,which is efficient and obtains good classification performance.To a certain extent,the method proposed in this paper narrowed the "semantic gap".