Abstract:Kansei image retrieval is a new kind of retrieval technology with high complexity. However, it's likely that only some parts of the image would attract people and produce affections. Color imposes a great impact upon the feeling as the basic feature of image, and the entropy of the image also exhibits the information quantity. In this paper, we present a method of kansei image retrieval utilizing the color and entropy to extract regions of interest (ROI). Back propagation neural network is employed to map the color and entropy of ROI to affective feature space. Finally, we show some experimental results of ROI extraction and kansei image retrieval based on interest.

Kansei image retrieval based on region of interest