GPU实现的椒盐噪声自适应中值滤波算法

提出了一种基于先检测、后滤波的椒盐噪声滤除GPU实现算法。该算法首先依据椒盐噪声的特征，将图像像素分为噪声和信号，然后对噪声像素作标准中值滤波，如果中值为信号，则用中值代替噪声像素灰度值；如果中值为噪声，则取中值邻域，如果中值邻域为信号，则用它代替噪声像素灰度值；否则扩大邻域窗口，重复算法。但随着图像规模的增大，在CPU上执行的时间显著增加。分析并利用图形处理器（GPU）的并行处理特征，并在GPU上实现了本文算法。实验结果表明所提出的算法较经典的自适应中值滤波算法有较好的去噪效果和边缘保护能力，并且随着图像规模的增大，算法执行时间较CPU可以提高最多3个数量级。

Improved Adaptive Filtering Algorithm Based on Programmable GPU

According to the characteristics of Salt-and-Pepper noise, an improved adaptive filtering algorithm is proposed in this paper. The principle of the method is Test Before Detect （TBD）. Accord- ing to the salt and pepper noise characteristics, the pixels of the image are divided into noise and signal, and then the noise pixel takes the standard median filter. If the value of median is a signal, replace pixel gray value with the median; if the value of median is a noise, then take the value of median neighbor- hood.If the median neighborhood is a signal, then replace the noise pixel gray value with it. Otherwise, neighborhood window is expanded. With the image size increasing, execution time on the CPU increas- es significantly. This article analyzes and uses parallel processing features of the graphics processor （GPU）, achieving the algorithm on GPU. The experimental results show that the proposed algorithm can not only effectively removes salt and pepper noise, preserves the edges, but also reduces the time consumption significantly.As the image size increases, it can increase up to 3 orders of magnitude com- pared with the CPU.