Studies On Related Technologies Of The Mapping Between Visual Features Of Images And Emotional Semantics

The image which is rich in emotional semantics can evoke different emotions of mankind. Image retrieve which uses the emotion as clue has become one of the forefront research topics. In the study we found that the visual characteristics of images and image semantic mapping problem are the core of this subject. It involves a multi-disciplinary field, which includes cognitive science, psychology, image processing, pattern recognition and artificial intelligence. The studying results can be applied to many aspects, such as advertising design, image retrieval, and psychological research. So it has drawn widespread attention in recent years.Due to the sophisticated, changeable, vague and loose relationship between image visual features and emotional semantics, when studying the image mapping, we need to address following problems: effective modeling of user emotion, matching image visual features with emotional semantics, extracting matching rules and so on.This dissertation focuses on a study of the following issues.(1) Proposed an effective computing model for a user to realize the user emotion modeling. The Model calculates the user's emotional intensity, emotional decay and factors among emotional interaction, and takes the theme-stimulating and the external environment as a parameter to join effective computing. Then the prototype of the effective computing systems is built. Take the 852 marked by the CAPS of Chinese Academy of Sciences as the annotation test data to test to validate this model, and experimental results presented in this dissertation show the accuracy of the computing model.(2) Proposed an effective semantic ontology library's construction method based on MPEG-7 and the fuzzy concept lattice which solves the matching problem between image visual features and emotional semantics. This dissertation defined the ontology frame based on the MPEG-7 description standard and constructd the effective semantic application ontology applying the fuzzy concept lattice generation algorithm. Using the ontology's semantic features and reasoning ability, we achieve the fuzzy emotions rule-matching algorithm to link up the image visual features and emotional semantics on the basis of establishment of a stable fuzzy concept lattices body restraint mechanism. We use the text annotation tool Protégé3.4 and reasoning tools RacerPro 2.0 to carry out the simulation experiments of ontology reasoning. The results show that the ontology library of image emotional semantic and related algorithms, which integrates the MPEG-7 and fuzzy concept lattice, achieve higher accuracy and better resolves the matching problem of image visual features and emotional semantic.(3) Proposed a rule simplification learning algorithm of image emotion semantic. Combined with rough set theory, this algorithm simplifies the attributes of image emotion semantic rule training sets. This algorithm uses decision tree learning algorithm to classify image emotion semantic rule training sets, and uses mini-max rule determine to simplify image emotion semantic rules, and the accuracy of the mapping rules between image visual features and emotion semantic. In this experiment, we select standard image texture and emotion vocabulary corresponded data set Texture Emotion and Lymph, Breast-cancer, Zoo data sets in UCI to conduct comparative experiments about rule learning. The results show that image emotion semantic rule simplification learning algorithm based on rough set and decision tree this dissertation proposed has higher accuracy and more reasonable reduction degree, and realizes mapping rule learning between image visual features and emotion semantic.(4) Based on the software requirements of emotion-based image retrieval in the field of printing advertising design, this dissertation proposed image emotion semantic retrieval system model with a core of ontology. According to this model, this dissertation adopts some related technologies we proposed to establish image retrieval system architecture based on emotion semantic, and design image retrieval system with image feature automatic annotation and based on emotion semantic in advertising field. This system demonstrates the value of this research.