Research On Image Emotionclassification Algorithm

Images can convey the rich emotion for the observer.In recent years,the image emotion classification has become a new research topic in the field of computer emotion analysis.As a high-level semantics,there is a "semantic gap" between emotion and the underlying characteristics of the image.How to find the emotional characteristics of the image,and how to establish emotional description mechanism of human perception have become the key to the realization of image emotion classification.This paper combined visual cognitive theory with psychology,pattern recognition and other fields of knowledge to study the emotional classification of images.First,because the perceptions at the perceptual level of human beings are often influenced by the significant regions in the image,and the high-level cognition makes people pay more attention to the facial expression and the specific target in the image,this paper proposes a Multi-feature fused image emotion classification method to improve the context-aware significant region detection.Enlarging the range of significant regional extraction,combining the human high-level cognitive extraction of the face area in the image,and forming the content area of image emotion,and finally performing multi-feature fusion and describing the classification process.The algorithm can filter out some non-emotional interference in the image and improve the accuracy of classification.Secondly,as for the fact that the human face always show up in the image screen,the position of facial expression classification in image emotion classification is improved,and the multi-classifier semantic description classification framework is constructed.According to whether human face is detected in the image and the different proportion of the face area,different classification will be processed,and emotional semantic description and classification of the image will be processed from the bottom up.The frame can significantly improve the accuracy of the image emotion recognition of the character image in the data set.Finally,an image emotion classification method based on Sentibank detection is proposed.The basic idea of this method is to convert the image into a text description,taking into account that the phrase itself also has different emotional classification,when the image is described in phrase sequence.By calculating the ANP 's text emotion value as the weight,the ANPs response is weighted,and the emotion value of the image is constructed to carry out the emotion prediction.Experiments show that compared with the common classification method based on ANP response,this method improves the classification accuracy.