边界噪声检测耦合差分曲率驱动的脉冲噪声图像降噪

Denoising Algorithm of Impulse Noise Image Based on Difference Curvature-driven Diffusion Model of Boundary Discriminative Noise Detection Coupling

目的提出边界识别噪声检测耦合差分曲率驱动扩散模型的脉冲噪声图像降噪算法,用于高密度(≥50%)脉冲噪声的消除。方法基于传统边界识别噪声检测BDND,定义噪声像素分类规则,设计新的边界识别噪声检测M-BDND机制;定位噪声边界,精确识别噪声像素点,形成噪声区域与完好区域;利用噪声像素点的周边信息形成掩码,对其进行修复,有效填补噪声像素点,而在完好区域只将像素点进行复制;构造了差分曲率驱动扩散模型控制噪声像素区域的扩散过程,完成图像复原,并对该模型进行了数值分析。结果与当前图像降噪技术相比,对于等密度随机值脉冲噪声而言,提出算法的误检率与虚警率更低;在噪声密度高达90%时,仍然具有可接受的降噪效果,能够更好地保留原图的边缘与细节特征,且复原图像的一维行距像更好,与真实图像的吻合程度高。结论该算法可用于高密度脉冲噪声的图像降噪处理。

The aim of this study was to propose a pulse noise image denoising algorithm of boundary discriminative noise detection coupling difference curvature-driven diffusion to remove high-density impulse noise （≥50%）. Based on traditional boundary discriminative noise detection, the noise classification criterion was defined and a modified BDND algorithm was designed to detect the M-BDND model to locate noise edge, accurately identify noise pixels and form noise domain and non-noise domain. Then the domain information of noise pixels was used to form a mask to effectively restore the noise pixels. It only copied the pixels from the observed image in non-noise domain. The diffusion process of noise-pixel domain was governed by difference curvature-driven diffusion model to accomplish image restoration and a numerical analysis was conducted for the difference curvature-driven diffusion model. The experiment results showed that the proposed algorithm had lower mistake rate compared with the current image denoising method for equidensite random impulse and had a good noise reduction effect and was more able to retain the edge and the detail feature of image even if the noise density was as high as 90%. The restoration image of one dimensional row spacing was better and had a high degree of agreement with the real image.