Perceptual Constrained Remote-sensing Image Compression Algorithm

The remote-sensing imaging has been the main method to observe the earth surface, and been utilized maturely in military and civil applications. With the development of electronic-imaging technology, the resolution and sample rate become higher and higher, thus more data needs to store and transmit, and this brings difficulties to transmit data from satellites to the earth and imag processing on satellites. To contrapose these problems, the image must be compressed effectively.The compression algorithms based on DCT transform would make block-artifacts significantly at lower bit-rate, and can't guarantee the perceptual quality as well as the mathematical fidelity of image. By importing the wavelet theory, some compression algorithms based on wavelet transform, such as EZW and SPIHT, bring renovation to the image compression field and meliorate the compression efficiency greatly. Although these algorithms exhibit good performance when compressing natural images; the obvious coding effects would take place, such as ringing, when compressing image with high resolution and rich texture at low bit rate.If only from the viewpoint of information theory, it is very difficult to guarantee the perceptual quality of the reconstructed image with rich texture, no matter what measure is adopted. However the human observer can tolerate some distortion, which changes with contrast, texture and the luminance of background. A locally adaptive wavelet image coder is presented in this paper. It utilizes an HVS (Human Visual System) model in wavelet domain and tunes the quantization step for each DWT coefficient adaptively by combining the space-and frequency-localization properties of wavelet decomposition and local properties of image. It can make the reconstructed image have optimized visual quality.The proposed algorithm has some salient properties: (1) In order to reduce the computational complexity, the lifting scheme Integer Wavelet Transform (IWT) of D9/7 is adopted. (2) The quantization step for every DWT coefficient is adjusted locally adaptive to optimize the perceptual quality of reconstructed image based on an HVS model. (3) The quantization step can be inferred from data decoded by taking into account the multiresolution structure of wavelet decomposition, so there is no need to send side information to the decoder. (4) The image is divided into small blocks which are encoded independently, and the compression ratio of each block is controlled by the statistical characteristics of the image to improve the fidelity over the whole image. Compared with JPEG2000, the proposed algorithm exhibits superior performance in terms of perceptual quality despite of the PSNR decreasing little. Experiment results illustrate our algorithm's validity.