A novel very-low-bit-rate textured image compression scheme is proposed in this scheme, both texture self-similarity property and affine transform are used as basis to achieve coding tasks. An input texture image can be reconstructed from both the prototype block spectral and the corresponding affine transform parameters. These parameters are estimated from the prototype block spectral and that of the rest blocks using the multiresolution Fourier transform, MFT, of the input image. In the synthesizing process, two different groups of parameters are generated, the MFT coefficients of the prototype block and the estimated affine parameters amongst blocks. In order to achieve the very-low-bit-rate coding scheme, this paper presents the quantization and coding scheme of the transformed coefficients and the estimated affine parameters. Consequently, figures of the compression and distortion ratio of the reconstructed image can be measured to analyze the structure of textured image over various scales.

A model of natural texture based on a structural component that uses affine coordinate transformations and a stochastic residual component is presented. It is argued that the selection of an appropriate analysis scale can be formulated in terms of a tradeoff between the rate at which parameters are generated and the distortion resulting from the approximation by the structural component. An efficient algorithm for identifying the parameters of the structural model is described and its utility demonstrated on a number of synthetic and natural textures.

A novel frequency domain approach to the analysis of texture is presented. It consists of two main components: texture synthesis and segmentation. The synthesis method models the relation between pairs of texture patches by an affine transform. Segmentation is accomplished by detecting the texture boundary using a gradient operator. This scheme is implemented in the Fourier domain with varying scales using a generalised wavelet transform, the multiresolution Fourier transform (MFT). The technique has the potential to deal effectively with textures having varying amounts of structural information.>