Research On Recognition Of Image Emotional Semantics Based On Feature Fusion

In recent years,with the continuous development of computer technology and emotional semantic research,computer emotion recognition has become increasingly widespread.The application of computer image sentiment analysis in the field of psychology has shown a gradual upward trend.In our lives,the image information caused by our human vision is the most direct and specific form.Therefore,visual information has a great impact on our emotions.Thousands of images,a large amount of image information brings great difficulties to people,and quickly and accurately discovering relevant information has become the focus of attention.Using computers to automatically analyze and understand images becomes an effective way.As a high-level emotional semantics,not only must the computer understand and understand the image content,but also achieve effective organization,classification and image management.Establishing the connection between the low-level and high-level semantic features of human perception and emotion mechanism becomes the key to realize image emotion classification and recognition.In the thesis,the three aspects of the problem are deeply studied,and the emotional recognition of images is realized.The main research contents and innovations of this paper are as follows:(1)Explain the low-level features,color,texture and shape features of the image.Select the typical extraction technology,and at the same time,the results are obtained,and the low-level feature space of the image is established.(2)The emotional space is proposed.Through the sentiment questionnaire method,the emotional semantic database is established,and the database is quantized to obtain the emotional space.(3)In terms of image sentiment semantic mapping,we first use the clustering-based SVM classification method and the clustering method to select the fuzzy C-means clustering algorithm(FCM),but the FCM-SVM algorithm is insufficient,and then improve the FCM-SVM algorithm and join Linear Discriminant Analysis(LDA)for more efficient classification performance.The paper comprehensively analyzes the low-level feature extraction technology and the image emotion semantic mapping mechanism,highlights the main algorithms of color,texture and shape feature extraction,and the high-level emotional semantic recognition module of the image,and clarifies the emotional modeling and emotional space establishment.From the normative point of view,the corresponding association between the underlying features of the image and the high-level emotional semantics is proposed,and the emotional semantic recognition mechanism is established.Combining various image underlying feature mapping methods and emotional semantic mapping methods,it is found that LFCM-SVM is the most suitable mapping method.