摘要
为快速地去除或减少DSA(Digital Subtraction Angiography)图像的噪声,对比评价KNN(K Nearest Neighbors)算法对高斯噪声、泊松噪声、斑点噪声、椒盐噪声4种噪声去除或减少的效果,帮助医生快速准确地为病人诊断疾病。提出的算法主要贡献在于构建了基于GPU(Graphics Processing Unit)的加速方法,使传统图像去噪的运算速度得到大幅提升。基于图像降质、图像还原过程建模,使用KNN算法对4种噪声去除或减少,并对算法做并行化处理,利用GPU加速实现去噪的过程。通过实验得出,KNN算法能较好地去除或减少高斯噪声、泊松噪声来还原DSA图像,使用CUDA(Compute Unified Device Architecture)编写可在GPU上运行的程序,利用GPU对1 024×1 024像素的24位深度的DSA图像去噪,平均渲染帧率能达到190.53 f/s(帧/秒),较传统CPU(Central Processing Unit)串行,平均处理速度提高70.86倍。使用GPU加速能够快速地处理数据量较大、计算密集的DSA噪声图像,实现有效并且快速的高斯噪声去除,帮助医生精、准、快地诊断疾病。
The purpose of this paper is to efficiently reduce or eliminate the noise of DSA ( Digital Subtraction Angiography) image and comparatively evaluate the efficiency of noise reduction of including Gaussian noise, Poisson noise, Speckle noise and Salt and Pepper noise based on KNN ( K Nearest Neighbors) algorithm. The resuhs can be used to provide high quality image to doctors for quick and accurate disease diagnosis for the patient. The main contribution of the proposed paper constructs an acceleration struc- ture using GPU, which can dramatically reduce the time cost of the traditional image denoising algorithm. The process modeling of image quality and image reconstruction is supplemented in KNN algorithm to reduce or eliminate the four kinds of noises. Addition- ally, the parallel processing based on GPU (Graphics Processing Unit) is executed to accelerate the denoising procedure. The ex- periments prove that Gaussian noise, Poisson noise can be removed or reduced effectivel~~ to reconstruct DSA image by KNN algo- rithm. The developed programming codes using CUDA (Compute Unified Device Architecture ) works well in GPU to realize the average rendering frame rate 190.53 f/s for denoising of 1 024 × 1 024 DSA image with 24 bit depth. Compare with the traditional CPU ( Central Processing Unit) serial, the average processing speed rate is increased by 70.86 times. Using GPU acceleration can quickly process DSA noise image with large amount of data and intensive computational mission. The results can provide accurate and fast diagnosis to doctors with the efficient Gaussian noise reduction.
出处
《电视技术》
北大核心
2016年第6期10-16,共7页
Video Engineering
基金
国家自然科学基金项目(61473112
61203160)
河北省教育厅项目(QN2014166)
河北省自然科学项目(F2015201196)