Features for image retrieval: The impression degree of a human image by overexposure occurring in the facial area

To improve the retrieval accuracy of content-based image retrieval systems, it is important to reduce the “semantic gap” between the visual features and the richness of human semantics. Overexposure in the facial area of an image, which makes parts of the face look unnaturally white, directly affects the impression of the image. If the impression of a human image can be automatically calculated on the basis of its content, an automatic indexing system focusing on human semantics can be developed. This can help to improve the efficiency of image retrieval and accuracy in searching images on the basis of facial expressions. In this study, we investigated the overexposure impression degree (OID), a feature that expresses the impression degree of a human image by overexposure. We also developed an algorithm for judging the OID. The algorithm includes three steps. First, it detects areas of the face, eyes, lips, and skin of a human in each image in a preprocessing phase. Next, it uses the color difference in the CIELAB color space and extracts overexposed areas from the face on the basis of color information on overexposure. Finally, the algorithm classifies an image into three types of OIDs on the basis of the extent of overexposure in prone areas of the face. The results of experiment conducted with 11 evaluators for analyzing images suggest that the proposed method can search and retrieve images that match the search queries with an F-measure of 94.7%.