Image Retrieval System Based On Individual Multi-Layer Affective Model

With the coming of information society, the technology of multimedia and intelligent information process developed rapidly. The uses of image become widely in each way, and the management and retrieval of image information become more important. Currently, how to manage and retrieval huge image databases effectively is a hot research direction. The image searching technology has gone through text-based image retrieval (TBIR) stage, content-based image retrieval (CBIR) stage, and semantic-based image retrieval stage which is developing nowadays. But the major Internet searching engine providers still use text-based image retrieval. The content-based image retrieval technology becomes the mainstream from 90s. However, people distinguish the difference of images not only through the visual characters, but also through the meaning of images which can't be got from the visual characters directly but from the understanding of images' contents. The understanding is semantic characters of images. None but combine the multi-character, especially semantic information, can the capability of retrieval system approach the human mentally level. The semantic-based image retrieval system is just a technology which search images based on the logic characters and abstract properties, and the technology is to improve the search ability of computer in order to approach the human's mentally level. The semantic-based image retrieval system is still in research and discussion stage. The affective-based image retrieval system is the one hotspot of the research.This paper's main research direction is the affective-based image retrieval system. Firstly, the existing image affective models and the main methods of feature extraction in image retrieval are introduced in details. Secondly, the group phenomenon of human emotion in image affective models is pointed out based on study and research of the existing image affective models and some mainstream views on the psychology of human emotion. Combined with this view, some improvements of the existing image affective models are done and a multi-layer individual affective model is introduced. This model divides emotion into two levels: the layer of public emotion and the layer of individual emotion. Public affective layer is the common interpretation of image emotional information. This paper notices the group phenomenon of public emotion, namely the impact on public emotion by group differences which produced by social attributes such as sex, age, people, nation etc, using prior knowledge, production rules are constructed. Then combining with the individual information provided by users when they registered, the group of users are determined, and the users' information of public emotion layer are constructed. The layer of individual emotion is established by use of the based-on user feedback method. The users' evaluation to the retrieve result or the active study is used to establish and improve this layer. Finally, in accordance with the multi-layer individual affective model, this paper establishes an image retrieval system, which has made good search results in testing.