Research On Several Key Issues Of Emotional Semantic Image Retrieval

With the rapid development of vision sensor technology, digital cameras, video cameras and other multimedia equipment have entered nearly every household. Under such circumstances, people's ability to produce, share and store iamge information has been greatly enhanced. The amount of image grows so fast that we can't find target image that we want. Image retrieval has became an active area. Now there are three kinds of primary technology for image retrieval: text based image retrieval,content based image retrieval and semantic based image retrieval. The former two methods are relatively mature, but they have inherent shortcomings, which can not help solve practical image retrieval problems. In order to overcome the adverse effects of semantic gaps, this paper uses semantic-based methods.Images contain rich semantics. As the high and important level semantics, the emotional semantics plays an important role in semantics retrieval research. Emotional semantic representation, image feature extraction and emotion recognition are the three key issues on this area.This paper makes a deep research to the three issues, proposes a new image representation framework, and produces a prototype system to recogize emotions of image. Based on the framework, the paper selects color,spectrum and lines as the visual features which could reflect image emotion semantics. Neural network, based on brain science, is the simulation and abstraction of brain. So the paper choose the BP neural network as the emotion classification to test the prototype, and proves that the method has good performance.Color, closely linked to emotions, has the power to arouse emotions. SVM, based on statistics science, displaying unique advantages in solving little sample problems, and non-linearity and high-dimensional patterns recognition problems, is becoming a newly active area in machine learning. In chapter four, the paper uses the SVM as the emotion recognizer and color mean value and color distribution as the features for emotion recognition. Experimental analysis on this method has been conducted in this paper and its effectiveness has been proved.At the end, major tasks and creative points have been discussed and suggestions for future research provided.