The Study Of Image Affective Retrieval Based On PAD And Fusion Of Multi-Features

With the rapid development of multimedia technology, the image information is increasingly enrich people’s lives, and how to use the computer retrieved ideal image quickly and accurately become a hot topic currently. Scholars have put forward many kinds of image retrieval methods, such as text-based image retrieval, content-based image retrieval method and the image retrieval based on image affective semantic classification, but they have their own focus so that they cannot be more complete expression of the information image contained. Therefore, in this paper, image affective retrieval based on PAD and fusion of multi-features method has been proposed, from the following three aspects:Firstly, design an image retrieval system based on low-level visual features and used overlapping block feature extraction algorithm. After experimental validation, this method can extract the visual characteristics of the image better, but the search results are poor when the image affective as the retrieve target.Second, in order to overcome the semantic gap between image visual features and affective information, the article proposed a method of neural networks. The method mapped visual features in continuous PAD emotional model to complete the prediction of image emotional information. After experimental test, the image affective semantic retrieval using this method gets better result. But seen from the search results, the visual similarity of the image is not satisfactory.Finally, after analyzed the two experimental problems comprehensively, this paper proposed a retrieval method closer to human visual perception image affective retrieval based on PAD and fusion of multi-features. The method is based on the good visual features and emotional access to information, using multiple linear regression method for information fusion of multi-class characteristics. In the fusion process, designing scientific behavioral experiments to obtain relevant data and make the feature fusion model more in line with the human visual mode. On this basis, verification system was achieved, and then analyzed the results obtained from repetitive experiments in the system. It proved that this method can combine the visual characteristics of the image and emotional information and improve the performance of the retrieval system, which has important theoretical and practical significance.