基于小波多尺度分解的Mean Shift图像滤波方法

将小波多尺度分解与传统Mean Shift滤波算法相结合提出的一种有效的图像滤波方法。先将含噪声图像进行Mallat塔式分解，获得不同尺度、不同频带的子图像。将低频近似图像保持不变，对高频细节进行Mean Shift滤波，最后将低频近似图像与高频滤波后的图像进行合成得到去噪后的图像。由于Mean Shift算法是一种迭代方法，要保证较高的数值计算精度则需要较多的迭代次数，耗费较长的计算时间，为克服这一缺点，提出了采用Fourier级数来近似计算高斯函数。实验结果表明该方法在降低噪声的同时能够尽可能的保留图像细节，其去噪效果优于传统的高斯滤波、Wiener滤波方法和单一小波域值法和Mean Shift滤波方法。

Image Mean Shift filtering method based on wavelet multi-resolution decomposition

Combined with wavelet transform and Mean Shift filtering,an efficient image filtering method is presented.Firstly,noised image is decomposed to multi-resolution sub-band images by Mallat pyramid decomposition.High frequency sub-band images are denoised by Mean Shift filter,without any changes to the low frequency approximation image.Denoised-image is obtained by composing the high frequency detail images with Mean Shift filtering and low frequency approximation image.However,Mean Shift is an iteration scheme.It wastes more time to calculate and needs more iteration to ensure higher numerical accuracy.In order to overcome its disadvantage,Gauss function is calculated approximately by Fourier series.Experiment result shows that the noise of the image is removed effectively.At the same time,the detail of the image is kept well.The method has better denoising effect than traditional Gauss filtering method,Wiener filtering method,single wavelet thresholding method and Mean Shift filtering method.