论文部分内容阅读
小波变换编码是目前研究较多的图像压缩方法 ,变换系数的量化是获得低比特率、高信噪比压缩图像的关键步骤。为了设计最优量化器 ,必须确定变换系数的分布规律。选择“Face”、“Girl”、“Lena”和“Panda”4幅标准图像数据进行统计研究 ,用长度L =1 8的Vetterli双正交小波将 2 5 6灰度级 2 5 6× 2 5 6图像分解为 3层1 0个子带 ,使用“KS”测试统计方法确定图像小波变换系数的分布规律。给出了瑞利分布、高斯分布和拉普拉斯分布假设下的“KS”测试统计结果。统计结果表明 ,低频部分符合高斯分布 ,其余部分符合拉普拉斯分布。模拟结果显示 ,假设高频部分符合拉普拉斯分布可获得比高斯分布假设更高的恢复图像信噪比。
Wavelet transform coding is the most commonly used method of image compression. Quantization of transform coefficients is a key step to obtain low bit-rate and high signal-to-noise ratio (S / N) image compression. In order to design the optimal quantizer, it is necessary to determine the distribution law of the transform coefficients. Four standard image data, “Face”, “Girl”, “Lena” and “Panda”, were selected for statistical research. Viterbi biorthogonal wavelets of length L = 18 were used to divide the 256 gray levels of 2 5 6 × 2 5 6 image is decomposed into three layers of 10 subbands, and the distribution of the image wavelet transform coefficients is determined by using the “KS” test statistical method. The statistical results of “KS” under Rayleigh distribution, Gaussian distribution and Laplace distribution are given. The statistical results show that the low frequency part is Gaussian and the rest is Laplacian. The simulation results show that the SNR of the recovered image is higher than that of the Gaussian distribution, assuming that the high frequency part conforms to the Laplace distribution.