Research Of Image Emotional Classification Algorithm Based On GEP

Along with the massive growth of internet image data and the rapid developmentof the Human-Computer Interaction System, the question about how to effectivelyclassify the image, which will make people to retrieve the desired image in the vastresources quickly, has increasingly get people’s attention. In recent years, the imageemotional semantic classification has become an active field of computer visionresearch. It has important guiding significance in helping Computer to correctlyexplain and perceive image content, to achieve effective image organization,classification and management, etc. The study of the image emotionalclassification,which is intended to better describe and explain the whole content ofthe image and achieve the effective management of image, relates to patternrecognition, affective computing, physiology, visual psychology and otherdisciplines,, has profound significance and great commercial value.In this thesis, we study the emotional image classification combined with theknowledge of visual cognition theory, psychology, pattern recognition and other fields.Because of GEP algorithm has efficient classification performance. we design andimplement image emotion classification algorithm based on GEP， Firstly, we useCSIFT algorithms which is an improved SIFT feature extraction algorithm to extractimage features, and get the color information and other related features with strongsemantic discrimination for emotional. Secondly, in order to generating a digitalimage matrix,we select the typical feature vectors that include emotional labels byusing k-means clustering algorithm, Finally, we design a new kind of chromosomeand fitness function, and implement image emotion classification algorithm based onGEP. The purpose of classifying the image emotion is achieved.We conducted classification experiment on eight kinds of emotion in imagelibrary which is from the two networks data sets: Flicr and Deviantart. The resultsshow that the algorithm has good performance in the image classification, andcompared with traditional SVM algorithm, its accuracy is improved by16.7%.