摘要
为了解决传统图像去噪算法存在边缘纹理信息损失的问题,根据图像平滑区域DCT非零系数个数较少的特点,提出了基于3DDCT的图像去噪及增强处理方法。该方法首先依据l2范式将含噪图像的相似区域块构成块群;再根据块群中块内像素的相关性,对各块进行2DDCT变换并利用阈值进行首次去噪,根据群内块间对应像素的相似性,对块群进行一维DDCT得到去噪且增强的图像。与传统算法相比,该算法在去噪过程中扩大高频非零系数进而增强边缘纹理信息,提高了图像的视觉效果。
To solve the problem of the loss of information in texture areas using traditional image denoising algorithms, we propose an algorithm for image denoising and enhancing based on 3DDCT according to the characteristic that the number of non-zero discrete Cosine transform （DCT） coefficients in smooth regions is fewer. First, similar blocks of the images is put to a block group according to l2 normal form; second, 2DDCT transformation is applied to each block and the threshold for the fin＇st denoising is used according to the correlation of pixels within the block; after that, 1DDCT transformation is applied on block groups according to the similarity between the corresponding pixels; then the threshold is used for a second time denoising; third, a power expansion is made to the non-zero coefficients in high frequency regions of 3D transform domain; f＇mally, the proceed image is combined with Kaiser Window function. Compared with traditional algorithms, the new algorithm has a better visual performance because of its abilities to enlarge non-zero DCT coefficients in high frequency areas and to enhance information in texture and edge areas at the same time.
出处
《电子科技大学学报》
EI
CAS
CSCD
北大核心
2011年第5期742-746,共5页
Journal of University of Electronic Science and Technology of China
基金
国家自然科学基金(60773168)
四川省科技支撑项目(2010GZ0169)
