Research On Emotional Attributes Of Digital Images Based On Deep Network

With the rise of mobile Internet,more and more users like to express their emotions by uploading real-time images on weibo,WeChat and other mobile social software.Observers can stimulate inner emotions after watching the uploaded image by the users,so the images have emotional attributes.The study of the emotional attributes of the digital images is of great significance to the analysis of the emotional state of the users.At present,most of the researche on the emotional attributes of digital images is based on the emotional classification task.The existing research methods usually judge emotional attributes of images by training classification models,but ignore the importance of emotional regions in the research of emotional attributes of images.Therefore,to explore emotional attributes of digital images,this paper starts from two directions of emotional regions and emotional classification of images.The innovation of this paper is as follows:(1)An algorithm for detecting the emotional regions of digital images based on saliency features is proposed.This paper first discusses the relationship between the emotional regions and the salient object regions in images,and proves that the emotional regions contains the salient object,i.e.,the emotional regions and the saliency regions are of high coincidence with each other,which is often ignored by the existing emotional regions detection algorithms.Therefore,to accurately detect the emotional regions,this paper designs an emotional regions detection method based on saliency features,where the proposed network adopts a symmetric structure of encoding and decoding,and adds the saliency regions feature map to the decoding network to assist the detection of emotional regions so that can automatically detects the emotional regions of the image through the network model training.The experimental results demonstrate the effectiveness of the regional saliency features in assisting the detection of emotion regions,and the superiority of the proposed model over the state-of-the-arts.(2)An algorithm of emotional regional detection and emotional classification of digital images based on a multi task depth network is proposed.Traditional algorithms can not make full use of the correlation between image emotional category and emotional regions to learn sharing characteristics.Therefore,this paper designs a digital image emotional regions detection and emotional classification algorithm based on a multi task depth network.The multi task network includes the image emotional regions detection task and image emotional classification task.The two tasks can improve the generalization ability of a single task through the training of the parameters of the sharing layer,thus improving the accuracy of the network.The experimental results show that the multi-task emotional regions detection method can accurately detect the emotional regions of images and the detection performance is superior to the general single task emotional regions detection methods and the existing saliency and object detection methods.In this paper emotional regions features are used to assist emotional classification.The experimental results show that the performance of the emotional classification network based on the characteristics of emotional regions is improved,which proves the importance of emotional regions in the research of emotional attributes of digital image.The paper has 33 pictures,9 tables and 77 references.