中文摘要：

基于整数小波提出一种用于自然景物图像去噪的快速递推广义交叉验证（FR-GCV）算法,首先对图像做整数小波变换（IWT）,计算小波图像中小波系数的分布概率,然后利用整数递推,降低了GCV函数计算复杂度。最后通过小波系数下采样和阈值上界限定进一步降低了算法复杂度。实验结果表明,FR-GCV算法对自然景物图像去噪耗时比GCV算法降低了90%以上。FR-GCV算法能够快速求取去噪最优阈值,在静止图像去噪领域具有较强的实际意义。

英文摘要：

A computationally efficient fast recurrent generalized cross validation（FR-GCV） algorithm for natural image denoising is presented based on integer wavelet transform. Firstly the integer grey-scale pixels are transformed into integer wavelet coefficient, And its distribution is calculated. Then the fast recurrent procedure based on integer wavelet coefficient is applied,which takes full advantage of the relevance between GCV functions under adjacent thresholds. In addition,the FR-GCV algorithm further reduces the computational complexity by coefficients sampling and threshold upper bound modification. The experimental result shows that the time coplexity of FR-GCV is 90% lower than that of GCV. The optimal threshold could be efficiently obtained by FR-GCV algorithm,which is high practical significance in natural iage denoising.