Sub-image Coding Acceleration And Texture Segmentation Method

This paper presents several new methods to solve problems in fractal image coding and texture segmentation which are widely investigated in the field of image processing. The details are as follows:Over the past few years, fractal image compression has experienced a rapid growth and interested by researchers all over the world because of its desirable properties such as fast decoding, resolution independence of decoded image and high compression ratio. However, the computation complexity of fractal image coding is extremely high due to considerable number of domain blocks have to be matched with many range blocks in the image to be encoded. In order to accelerate fractal image coding, a novel algorithm is proposed based on detail information classification (DIC) and structure similarity (SSIM) index. D.IC is a new classification method given by us for the first time, which classified domain pool according to the distribution of gray detail information of the image block, while SSIM is a existing concept of image quality measure. There are two reasons to introduce SSIM to fractal image coding, on the one hand, SSIM not only consider luminance and contrast between image blocks, but also structure information comparing with conventional matching method, on the other hand, the time consuming of SSIM is half of the traditional one. Experimental results illustrated that the mixture of DIC and SSIM can save coding time greatly and obtain satisfactory decoding quality in contrast to traditional method. The concrete data are given below. For lena image, when the number of class is 30 and ε = 0.8, decoding speed is enhanced 64.48 times and decoding quality lose 0.95db compared with traditional method of fractal image coding. Furthermore, when rotation and reflection is not considered, coding speed is enhanced 505.54 times, and decoding quality loses 3.23db.Segmentation of texture image addresses the problem of identifying differentregions of homogeneous textural properties within the image. So the key technique is to extract a set of parameters that can label the characteristic signature for every textural class. In this paper, Multiscale Autoregreesive Moving Average (MARMA) model is introduced, and the coefficients of the model are treated as characteristic vectors. The starting point of develop MARMA model is establish the image sequence by transformation of Haar wavelet, then parameters of model is estimated according to niche genetic algorithms. Because multiscale analysis can obtain information of various scales, so the proposed method can be gain improved quality of segmentation. This method can deal with ordinary texture images and satellitic images with complex textural information.