Abstract: Aims at overcoming the defects of existing filters in noise detection and noise removal, this paper proposes an adaptive weighted median filtering algorithm based on detection with trimmed median. This filter first performs noise detection by using the minimum and maximum intensities, namely 0 and 255, which makes full use of the intensity characteristics of impulse noise, and then performs further noise detection based on the correlation and the approximation in intensity among neighboring pixels. And on the basis that the correlation among neighboring pixels decreases with the increasing of distance, different weighted coefficients are given to the noise free pixels at different distance in the neighborhood, the weighted median is used to remove noise. The size of neighborhood for denoising varies adaptively along with noise densities. Simulation results and analysis prove that the proposed method is capable of denoising effectively. Its denoising performance has outperformed significantly the existing distinguished filters in terms of peak signal to noise ratio, image enhancement factor and visual representation, especially for removing high density noise, it shows significant superiority over other filters.

Key words: weighted median filter, noise detection, trimmed median, minimum and maximum intensities, weighted coefficient, correlation

摘要： 针对现有算法在噪声检测与噪声滤除性能上的缺陷，提出修剪中值检测的自适应加权中值滤波算法。算法利用噪声的灰度特征，根据灰度最值0和255检测噪声，再根据邻域像素的相关性以及在灰度上的近似性，做进一步的噪声检测。根据邻域像素之间的相关性随距离的增大而减小的特性，对邻域中的信号像素分别赋予不同的加权系数，取加权中值以滤除噪声。算法去噪的邻域大小，随噪声密度和分布自适应地变化。通过去噪图像的主观视觉效果以及客观的去噪性能指标PSNR（peak signal to noise ratio）和IEF（image enhancement factor），仿真实验证明，所提出的算法相对于现有的算法具有更好的去噪性能，特别对于滤除高密度噪声，具有显著的优越性。

关键词: 加权中值滤波, 噪声检测, 修剪中值, 灰度最值, 加权系数, 相关性