Improved Generalized Co-occurrence Matrix for Image Retrieval using MPEG-7 Texture Edge

To achieve efficient image retrieval, this paper applied the five texture edge directions, including 0deg, 45deg, 90deg, 135degand nondirection, recommended by MPEG-7 standards to generalized co-occurrence matrix method. The new method firstly constructs five generalized co-occurrence matrices by computing the neighboring average of pixels in a fixed window in five directions respectively. And then it defines the mean values and variances of the texture features extracted from the five co-occurrence matrices as texture parameters for image retrieval. This approach take into account not only the direction information but also the correlation information between pixels, which integrates the advantages of traditional gray level co-occurrence matrix and generalized co-occurrence matrix. Experiments show that this new method obtains an advanced general performance to no matter gray images or color images comparing to the other two methods.