A Joint Texture Description Method Utilizing Visual and Semantic Features

Image texture is an important feature in content-based image retrieval system. To characterize the texture feature of images, we propose an effective texture description combining the visual and semantic features. It captures the visual feature of the texture in a greatly reduced texture spectrum scheme; furthermore, it can describe the semantic feature of texture in natural language thanks to linguistic variable. We also put forward a semantic feature extraction algorithm using neural network. Our experimental results demonstrate that the texture description has excellent performance in catching the visual and semantic content of the image texture. In some extent it can bridge the "semantic gap" between the low-level visual feature and high-level semantic feature in content-based image retrieval.