太赫兹成像NLM去噪算法优化研究

Research on optimization of NLM denoising algorithm for terahertz imaging

太赫兹透射式成像具有对人体无害,对非金属材料有较高的穿透性等特点。但成像图片信噪比较低,因此文章提出一种改进的非局部均值滤波算法,该算法先利用结构张量对图像进行分区处理。对平坦区域进行中值滤波,对非平坦区域采用改进的非局部均值滤波算法,该算法引入结构相似指标,图像通用质量指标,特征相似指标的高斯加权距离作为图像相似准则。为探索该算法在太赫兹透射成像图片的效果,文中根据太赫兹探测器类型的不同分为单点、线阵、面阵扫描成像实验,并展开成像研究发现线阵相机中存在坏点导致成像图片存在暗条纹,可利用高斯加权中值的方式消除暗条纹。单点扫描成像和面阵扫描成像由于运动平台移动因素和像元响应不一致性的干扰,成像图片具有噪声。将成像图片分别运用不同的去噪算法进行降噪重建,结果表明,与传统的非局部均值滤波算法相比,改进的非局部均值滤波算法对太赫兹透射成像图像处理后所得的峰值信噪比值有1~9 dB的提升,且和其他传统的去噪算法比较发现改进的非局部均值滤波算法处理后图像的峰值信噪比值也有2~8 dB的提升,去噪效果较好。

Terahertz transmission imaging has the characteristics of harmless to the human body and high penetration to non-metallic materials.However,the signal-to-noise of imaging pictures is relatively low,so this paper proposes an improved non-local mean filtering algorithm,which first uses structural components to partition the image.The median filtering algorithm is performed for flat areas,and the improved non-local mean filtering algorithm is adopted for non-flat areas,which introduces structural similarity index,image general quality index,and Gaussian weighted distance of feature similarity index as image similarity criteria.In order to explore the effect of the algorithm on terahertz transmission imaging pictures,this paper is divided into single-point,linear array and area scan imaging experiments according to the different types of terahertz detectors,and the imaging results show that there are dead pixels in the linear scan camera that lead to dark stripes in the imaging picture,and the Gaussian-weighted median can be used to eliminate the dark fringes.Single point scanning imaging and area scan imaging have noise due to interference from motion platform movement factors and inconsistent cell response.The results show that compared with the traditional non-local mean filtering algorithm,the improved non-local mean filtering algorithm improves the peak signal-to-noise ratio obtained after terahertz transmission imaging image processing by 1-9 dB,and compares with other traditional denoising algorithms and finds that the peak signal-to-noise ratio of the image after processing the improved non-local mean filtering algorithm also increases by 2-8 dB,and the denoising effect is better.