贺兰山岩画的多尺度非局部滤波算法

针对贺兰山岩画提出了一种新的滤波算法,该方法基于混合多尺度和非局部平均滤波的思想处理噪声图像.首先,将RGB彩色空间转换到L*a*b*颜色空间;其次,对L*a*b*颜色空间的每个分量进行多层小波变换;然后通过2种不同的策略处理粗尺度的小波系数,即对低频系数进行非局部平均滤波,对高频系数进行阈值处理,并对处理后的粗尺度小波系数进行重构得到上一层的低频图像;之后,对每一个尺度继续上面的操作直到得到最细尺度的系数,并对完全重构的图像进行非局部平均滤波;最后,将处理结果转换到通常的RGB彩色空间.大量的实验用于讨论参数的选取和算法的有效性.结果表明,该方法在计算效率、鲁棒性和视觉效果方面均优于已有的混合高斯尺度方法、多尺度双边滤波方法、非局部平均滤波方法.

Multi-scale and Non-local Filtering Algorithm of Helan Mountain Rock Arts

The purpose of the paper is to construct a novel kind of filtering algorithm to restrain the noise present in the images of Helan mountain rock arts.The approach deals with noisy image in L＊ a＊ b＊ spaces by hybridizing the concept of both multi-scale and non-local mean（NLM） filtering.Firstly,the RGB color spaces are transformed to L＊ a＊ b＊ spaces; Secondly,the wavelet transform is applied to every components of image,then the wavelet coefficients in the coarsest scale are processed in two different manors,specifically,the low-frequency（LF） is filtered by NLM and the high-frequency（HF） is addressed by shrinkage.After the LF and HF component in the coarsest scale are filtered,they are combined to form the LF component in the next finer scale and the similar two-manor processing is performed in this finer scale.The image is asymptotically reconstructed and the above process is applied to the wavelet coefficients in each scale until the finest scale is obtained,then the image in the finest scale （namely,in the spatial domain） is filtered merely by NLM.Finally,the L＊ a＊ b＊ color spaces are transformed to the common RGB spaces.Extensive experiments are conducted to discuss the choice of parameters and the effectiveness of the algorithm.The numerical results indicate that the proposed approach outperforms the three existing filtering algorithms（Scale mixtures of Gaussians method,Multi-scale bilateral filtering,Non-local mean filtering） in terms of computational efficiency,robustness,as well as visual effect.