基于人类视觉特性的两步去噪模型  

Two-step denoising model based on human visual property

作　　者：舒玉强[1] 杨红雨[1] 

机构地区：[1]四川大学计算机学院,成都610064

出　　处：《计算机应用》2009年第12期3372-3374,共3页journal of Computer Applications

基　　金：国家自然科学基金资助项目(60736046)

摘　　要：基于局部归一化的局部方差去噪模型在平坦区域去噪明显,但去噪速度慢,而且纹理细节保护欠佳。为此,用全局归一化的局部方差度量空间细节,得到新的扩散系数。新模型先用新扩散系数加速去噪,再用旧扩散系数去除不平整现象和小黑点。实验结果表明:新模型在快速有效地去除高斯噪声的同时可更好地保护纹理细节,而且信噪比和峰值信噪比与仅用一个扩散系数的模型相比有所提高。Removal model based on local variance normalized can remove noise significantly in fiat area but slowly and maintains the texture details poorly. Therefore, global variance normalized was proposed to measure spatial detail, and then a new diffusion coefficient was given. New removal model used the new diffusion coefficient to accelerate denoising and then used the old diffusion coefficient to remove the uneven and black dots. Experimental results show that new model not only removes noise quickly and effectively while protecting texture details better, and the SNR and PSNR are slightly higher than the model with only one diffusion coefficient.

关 键 词：人类视觉特性 直方图 扩散系数 空间细节 方差 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]