PERCEPTUAL TEXTURE SPACE FOR CONTENT-BASED IMAGE RETRIEVAL

摘要

Texture is an important visual feature for content-based image retrieval. An ideal content-based retrieval system should compare images in its database with the query in a manner that is consistent with human's perception of visual similarity. In practice, however, texture similarity computed using computational texture features is not necessarily consistent with human's perception. Although several perceptual texture spaces that capture human's perception of texture exist, the mapping from computational features to these perceptual spaces has not been derived. Another difficulty with using these perceptual spaces is that their measurements of texture similarity are confounded by texture scale and orientation because the textures used to construct the spaces are not normalized to canonical scales and orientations. This paper presents a perceptual texture space constructed using textures with canonical scales, orientations, intensity, and contrast. To illustrates the application of this perceptual space, computational texture features are extracted using Gabor filters and a multi-layer feedforward neural network is trained to map the Gabor features to the perceptual space. Test results show that the neural network can perform the mapping quite accurately. Moreover, retrieval tests show that retrieval performance is improved by mapping computational features to perceptual texture space.