血管内超声图像斑点模拟与滤波方法

Simulating and filtering speckle in intravascular ultrasound images

目的:在血管内超声(Intravascular Ultrasound,IVUS)中存在的斑点,严重影响图像的质量、图像的细节,以及图像处理的后续工作。为了研究IVUS图像中斑点的特性,以及验证IVUS图像去噪算法的性能,需要对标准图进行模拟加噪。在此基础上,为了提高IVUS图像的质量,便于图像的后处理,需要对IVUS图像进行滤波,在降低噪声的同时保留图像边缘,图像细节等的诊断信息。方法:采用Rayleigh噪声斑点模拟法,对标准图加入Rayleigh噪声后进行下采样,再利用Lanczos核函数进行差值得到所需的模拟图像,通过比较加噪图像与原图的相关系数和互信息,获得最优的模拟参数。基于双边滤波器,本文引入了一种改进双边滤波器算法,通过修改像素邻近度权值,以及像素相似度权值衰减程度的参数,自适应滤除乘性噪声,并且结合迭代,进一步提高滤波效果。通过比较滤波前后的PSNR以及SSIM,确定算法中的最优参数。结果:利用Rayleigh噪声,采样间隔为1或2,插值函数选用Lanczo2或Lanczo3,所得到的模拟图像与原IVUS图像的相似度最佳。对于改进双边滤波器σr取0.6,掩模大小取9×9,迭代次数取5可以取得最佳的滤波效果。结论:本文提出的Rayleigh噪声斑点模拟法可以有效地模拟IVUS图像中的斑点噪声。与传统的非线性滤波器相比较,改进的双边滤波器可以有效去处斑点,保护图像中重要的细节,本文所提出的改进双边滤波器达到了更为理想的滤波效果。

Objective The speckle in intravascular ultrasound （IVUS） has seriously affected the image quality, details, and the post-processing. Noise need to be simulated in the standard images to study the properties of the speckle in IVUS images and effectiveness of IVUS image danoising algorithm. And based on that, in order to improve the quality of IVUS images, convenient for the post-processing, the IVUS images need to be filtered, reducing noise but preserving edges, image details and other diagnostic message. Methods Rayleigh speckle noise simulation method was utilized to simulate the speckle noise, sampling after the the Rayleigh speckle noise was simulated in the standard image, and then obtaining the simulate image by the interpolation of lanczos kernel function. The optional simulation coefficients were obtained by respectively comparing the correlation coefficient and mutual information of the images with noise and the original images. Base on the bilateral filter, an improved bilateral filter algorithm was applied to remove the speckle noise more effectively by modifying the weight of the pixel proximity and the parameter which determined the attenuation degree of the weight of pixel similarity. Moreover, the filtering effect was further improved by combining with iteration. The optimal parameter was determined by respectively comparing PSNR and SSIM before and after the filtering. Results By applying Rayleigh noise, the simulate image, with 1 or 2 sampling interval, lanczo2 or lanczo3 interpolation function, had the highest similarity with original IVUS image. And the best filtering result could be obtained by the improved bilateral filter taking 0.6 for sigmar, 9＊9 for the mask, and taking five iterations. Conclusion The Rayleigh noise simulation method can effectively simulate the IVUS speckle noise in images. Compared with the traditional nonlinear filter, the improved bilateral filter can effectively remove speckle, protect important details in images, and achieve better filtering results.