Research On Rules Extraction Of Image Emotion Based On Immune Programming Algorithm

Color images usually contain much high-level semantic information. There is a certain corresponding relationship between affective features and low-level visual features of the image, such as color, texture, shape and so on. Revealing the emotion characteristics of images can not only describe images more effectively, but also be used in the area of personality image retrieval. But it is hard to build models for nonlinear mapping between the visual features and high-level emotions. Therefore, it is a recognized problem in the field of artificial intelligence by using computer to imitate human emotions and obtain accurate understanding of images emotional meaning.In recent years, the researchers spent a lot of energy on the relevant issues. The main method is about introducing artificial intelligence and machine learning techniques combined with classification or clustering algorithm, or using the methods of Human interaction, machine learning and by using external information sources. The research on Image emotion rules extraction is focused on building a description by using the basic characteristics of images for perception, recognition, and emotion information contained in the image features.To achieve the aim of using the form of better intelligibility rules to express characteristics of image information that contains emotion information, we must choose appropriate neural network to learn the mapping relationship between affective features and low-level visual features of the image.The research is related to computer science, artificial intelligence, machine learning and cognitive science, and many other disciplines. Because of the complexity of human emotion is, fuzzy and some uncertainty, many researches focus on the research of image are mostly limited on the identification and classification of several typical feelings. It is far from the ability of computer that understands and distinguishes human beings produces kinds of emotions in the real world. Therefore, it is necessary to extract rules from images.For the above, this paper made the following aspects of the research work. Firstly, it analyses the image low-level features which influence people's feelings.Secondly, it picks out landscape pictures from China Affective Picture System (CAPS) which marked emotional values for each pictures, and finishes the non-uniform quantization of the color features in its HSV color space. Thirdly, it trains the BP neural network by using color features and emotional values.Finally, it proposes a new neural network rules extraction algorithm based on Immune Programming. The algorithm based on the size of the fitness function, updates the population continuously to obtain the optimal clustering effect of hidden unit activation values, then obtains simple, high precision and good intelligibility rules by listing to form the relationship between the input and output neurons.The algorithm is used to establish low-level image features to emotion mapping, the results of experiment show it has high accuracy and efficiency.