基于局部特征的正则化滤波算法  

Regularization filtering algorithm based on local characteristics

作　　者：杨昊[1,2] 陈雷霆[1,2] 邱航[1,2] Yang Hao Chen Leiting Qiu Hang(College of Computer Science ＆ Engineering, University of Electronic Science ＆ Technology of China, Chengdu 611731, China Sichuan Provincial Key Laboratory of Digital Media, Chengdu 611731, China)

机构地区：[1]电子科技大学计算机科学与工程学院,成都611731 [2]数字媒体技术四川省重点实验室,成都611731

出　　处：《计算机应用研究》2017年第9期2817-2821,共5页Application Research of Computers

基　　金：国家自然科学基金资助项目(61202255);国家教育部广东省产学研重大专项项目(2012A090300001)

摘　　要：大部分针对脉冲噪声的去噪算法主要利用预设的全局阈值来对含噪图像进行恢复,因此对图像局部边缘和细节的恢复能力较差。对此,提出一种基于图像边缘特征的正则化滤波算法。算法首先通过反映图像边缘细节的局部统计量建立自适应阈值,用于进行脉冲噪声检测,得到噪声候选集合;然后以此为基础,优化保边正则化方法,提升算法性能。仿真实验从定量数值比较和主观视觉比较两个方面表明,该算法在噪声密度较高时,仍能提供较好的图像恢复效果,从而为图像的后续研究提供良好的基础。Most algorithms for removing impulse noise adopt the predefined global threshold to restore the corrupted images, therefore, they are not good at the recovery of local edges and details in images. Thus, this paper provided a regularization fil- tering algorithm based on edges, characteristics. The algorithm built an adaptive threshold by the local statistics which could reflect the edges and details in images. It adopted these thresholds to detect noisy pixels and got a candidate set filled of noise. Then it used the candidate set to optimize the edge-preserving regularization method for its performance. Simulation results show that proposed algorithm still can provide better performance of restoration by quantitative comparison and subjective visual comparison when noise level is high, thereby the algorithm can provide a good base for further research.

关 键 词：脉冲噪声 自适应阈值 保边正则化 滤波 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]