GPU实现的抑制椒盐噪声的排序滤波算法

Salt and Pepper Noise Sorting Filter Algorithm on GPU

文中提出了一种简单有效的排序滤波算法,并在GPU上实现了该算法。算法首先检测图像中的信号,如果检测的像素是信号则保留不变,否则取其邻域并对邻域内的非噪声像素排序并取中值,若中值为信号,则用它代替噪声像素灰度值;否则扩大邻域窗口,重复算法。随着图像规模的增大,在CPU上执行的时间显著增加。文中在CUDA平台上对其进行了实现。实验结果表明该算法不仅能有效地移除椒盐噪声,而且执行效率高,图像的规模越大,加速比越高,最多可提高3个数量级。

A novel sorting algorithm is proposed based on GPU. Firstly, according to characteristics of salt-and-pepper noise, one pixel to be processed is estimated to be a signal pixel or a noise pixel, the signal pixels are kept, otherwise the signal pixels of neighbourhood are sorting. Secondly, the median value is obtained from sorting data. The pixel value is substituted by median value if it is useful signal, or the area of neighborhood are expanded and repeat the previous step. In experimenting find out that the larger size of image is, the more time is on CPU. Compared with on CPU, time of parallel execution would greatly cut on GPU. The proposed algorithm which is used to denoise salt-and-pepper noise on GPU has the characters of the simple and efficient. Meanwhile, speedup ratio can be improved three or- ders of magnitude.