A Method Of Affective Annotation For Image Based On Fuzzy Approximate Reasoning

Affective computing which is involved in philosophy, psychology, aesthetics and anthropology, etc, is a novel challengeable research region. Atpresent, the discussion of image retrieval about affective computing has emergedfrom behind the research of affective computing theory and application. Somepapers mainly focus on the affective behavior of robot and wearable computinghave been published. And the same time, for more than two or three yearsquery-by-multi-example has been a popular query system for content-basedimage retrieval (CBIR). Although the two research regions have gotten up lately, some achievement have been gained, especially in affective computing research.Recent research has shown that the visual features of an image, such ascolor, texture and shape, play an important role in the content-based imageretrieval (CBIR). Image affecting people affective feeling not only depend onlow-level features such as color and texture; some objects in high-level imagealso give people different affective reaction. The dog in the image may makeone feel comfortable and flowers may make one feel well. Same object may make different affective effect, one feel kind for the water as it is source of lifebut afraid of flood. This is the difference form general image semantic retrieval.Combination of high-level semantic features and low-level visual featuresshould be highlighted in affective semantic classification, but implementation ofthe combination is difficult.Recently, there emerges research that considers computing emotion notwith mathematics method but psychology established by psychologists. HugoLiu developed a method to recognize the affective structure of a text document.In China, Zhang Ying made great efforts for recognizing and expressing affect ofhuman being, and a framework that extracts emotion from images is proposed,the authors point out that there is something important relationship between lowlevel features of images and emotion, such as color, texture and shape. Wangmade an image retrieval system that can be used for querying affective images,by syncretizing color and shape features. However, most of the papers aresummarized or only focus on the retrieval system framework, they do not careabout the annotation method.Based on the problem made mention of, in this paper, a framework ofimage retrieval system based on affective is proposed. We make some work liskthat emphatically: firstly, we select and modify algorithms that certified by agreat deal of test and reference documents for extracting visual features (color,texture and shape) from images, at the same time, we accumulate a lot oftranscendent expert knowledge. Secondly, we establish a proper emotion space and express style for quantifying the affective concept; we propose a fuzzyapproximately reason method to annotate the emotional images. And finally, inthe procedure of the expression method of extraction of visual features, wereference the MPEG-7, establish the MPEG-7 DDL, use the MPEG-7descriptors to express and store the visual features, and we use the ontology toexpress the emotion, design a component named L-T-H Linker (based on fuzzyapproximately reasoning method) which can link the low-level visual featuresto the high-level emotion ontology annotation.Our approach achieves a higher degree of precision, by asking 20 personsto estimate the emotion of 100 nature images and 100 decoration images incontrast with the system.