|
With the rapid development of Human Computer Interaction system, image emotional information has received more and more attention in resent years, and has already been the one of most hottest topic in man-machine interaction. The research shows that a image includes plentiful emotional information, and different images involve different emotions. Our purpose is to effective describe and express the emotional information through the computer. Thus, learning the rule of mapping which between image low-level feature to emotional feature is new and very challenge front topic in the field of emotion recognition.Image low-level feature has color feature,texture feature and shape feature, and all of them may produce different emotiom. Additonal,one person under the different state of mind or different time may produce the different emotion.The result due to ambiguity and uncertainty of emotion, and will hard to describe the emotion of the image.Paper introduct EEG feature according to these. EEG feature is a biometric feature which can directly reflect the thought,and have the advantages of easy collection,mature analysis technology and little user burden.Combining image low-level feature to accomplish the mapping between image low-level feature and image emotional feature.We make the work list that emphatically:Firstly, we accumulate the corresponding relations between the emotion and the image from a number of experiments, research and the references, and study the most common extraction of color and texture.Secondly, according to the advantages of mature technology and easy collection,collect the EEG when user enjoy the images, and the emotional vector is obtained under image stimulate from calculating the emotional matrix.Thirdly, according to the advantages of human brain and function analogic BP neural network,the input is image low-level feature and output is emotional vector, then train robust BP neural network and construct the rules of mapping between low-level feature and emotional semantic, at last, realize the mapping from low-level feature to image emotional feature.Finally, we select five hundreds of images from Corel database, testify the feasibility of BP neural network, at the same time, develop a retrieval system based on SOM, validate the result after the mapping, and analysis the experiment results.
|