Study Of Image Content Annotation Based On Emotion Semantic

Many of the images in the application not only bring a lot of representation of information, but also carried lots of emotion information. The current image annotation and retrieval techniques have mostly ignored the emotion factors. How to effectively express and describe the emotion images, and quantified, then reflect the user's subjective tendencies in the retrieval and annotation process, meet the user's emotion needs, will be an important and challenge topics in the field of image annotation and retrieval.Images contain rich semantics. As the very important level semantics, the emotional semantics plays an important role in semantics annotation and retrieval research. Emotion semantic representation, image feature extraction and emotion recognition are the three key issues in this area.This dissertation focuses on the emotional-based image annotation,and combined with game scenes take Chinese Academy of Sciences Institute of Psychology image library as the data source,and have researched the important technologies and algorithms on the emotional semantic annotation and retrieval of these data sources.Based on the initial classification of the image database, various visual emotional features of images in training database are extracted, and the application of SVM (Support Vector Machine) in the hierarchical emotional semantic classification and annotation is presented. Then certain SVM parameters and normalized feature vectors are used to train the SVM classifier, of which the influence to the emotional semantic classification and annotation is analyzed based on precision. The result of emotional semantic classification is an emotional semantic determinant tree, certain emotional semantic rules are gained for every emotional semantic classifier, and every emotional semantic concept in the database can be determined by some emotional semantic rules. The results show that appropriate visual feature and SVM parameter are helpful for the correct emotional semantic classification and annotation of image database.Currently, statistical learning theory and support vector machine SVM showed good results in the research with many ways, for the sample study can reduce the burden on users, applied to the relevant circumstances of emotional image annotation and retrieval better. 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 happiness degree and activation degree of positive and negative emotions as the features for emotion annotation and recognition. After the sample determined by the training model of SVM, SVM can be set to test the emotional identification. In this study, a total of 178 test images, the underlying characteristics of these images enter the SVM, can get their category. With pre-defined categories of comparison, according to the formula emotion recognition accuracy can be the appropriate classification and annotation of prediction accuracy. The results can be obtained by: selecting a suitable sample of the normalized prediction parameters can improve the classification and annotation accuracy, while the optimal parameter combination of the sample normalized experimental effect can be better.Overall, the paper mainly from the following aspects of emotion semantic image annotation and retrieval taken further, select the appropriate adjectives describing emotions feelings semantic image, select the two highly correlated with emotional image attributes, then this based on the use of a support vector machine SVM algorithm as emotion recognition algorithms, and the proposed method and the parameters for a large number of experimental results show that the method and parameters within a certain range of effective realization of the emotional image classification and annotation.