Image Retrieval by Emotional Semantics: A Study of Emotional Space and Feature Extraction

Image emotional semantic research is a promising and challenging issue. This paper analyzes the emotional space and builds a novel scheme that can annotate the image emotion semantic automatically and realize emotional image retrieval. Based on psychological experiments evaluating evoked feelings for art paintings, we first identify an orthogonal three-dimension emotional factor space of images through 12 pairs of emotional words. Then, the following three novel image features are designed for each emotional factor to predict it. They are luminance-warm-cool fuzzy histogram, saturation-warm-cool fuzzy histogram integrated with color contrast and luminance contrast integrated with edge sharpness. The values of emotional factors can be predicted from the image features automatically by using support vector machine of regression. Finally, we design and implement an emotion-based image retrieval system, which enables the users to perform retrieval using emotional semantic words. Experimental results show the effectiveness of our model.