Abstract: Visual retrieval by content in an Image DataBase (IDB)is still an open problem. So far, various methods withdifferent semantic levels have been developed, forinternet search or off-line IDBs, but few of them takeinto account the user's perceptual point of view. Twofeatures primarily used for visual retrieval in IDBsare shape and color. We focus our attention on color,from the perspective of color appearance. The HumanVisual System (HVS) has adaptation mechanisms thatcause the user to perceive the relative chromaticity ofan area, rather than its absolute color. In addition,due to the acquisition process, color distortions areadded to heterogeneous IDBs. Digital pictures of realobjects for IDBs must be digitized and the acquisitionprocess is composed of various passages and means, eachone introducing unwanted color shifting. Moreover, thecolor quantization and the device gamut can introduceadditional distortion on the original colorinformation. The overall result is a recognizable andthe device gamut can introduce additional distortion onthe original color information. The overall result is adigital image that can significantly differ in colorfrom the real object. For the user the image may stillbe easily recognizable, but the color search change canvary widely and differ for each image or for the sameimage with different acquisition processes. For thisreason, the user's perceptual point of view must beadded into the management of color. The idea presentedin this paper adds a pre-filtering algorithm thatsimulates the HVS and that discounts the acquisitioncolor distortion in the query image as well as in eachimage in the IDB. Moreover, we suggest to use for theimage retrieve, a more perceptively linear chromaticdistance in the color comparison. !15
展开▼