摘要
针对传统医学影像复原方法会产生丢失细节、边界模糊等复原效果不理想以及算法计算复杂的问题,本文采用了一种有效的全变分正则化迭代方法对含有模糊和噪声的退化医学影像进行处理。该方法结合ROF模型具有的保持边缘和结构的特性,充分利用图像能量间的梯度关系,在经典变分去噪模型中加入模糊核算子,对于该凸泛函模型采用变量分离的思想,引入二次惩罚项和松弛变量将图像变分复原的无约束优化问题分解为一系列子问题,结合交替Split Bregman技术和解的框式制约,直接对泛函进行迭代,同时引入阈值算子和收缩技术来优化子问题的求解,同时达到了保持医学影像重要边缘和细节信息和克服传统方法计算复杂的目的。仿真结果表明,与传统复原方法相比,该方法提高了图像的信噪比,均方误差也明显减小,克服了振铃效应,改善了图像的视觉效果,证明了该方法的有效性,且该方法具有良好的稳定性和快速的收敛性,可以更有效的应用于临床诊断以及后续分割。
The existing medical image restoration methods have some problems such as loss of detail, blurred boundary and algorithm computational complexity. Herein, we adopt an efficient total variation regularized iterative method to deal with blurred and noisy degraded medical images. This proposed method combines the characteristics of ROF model to preserve edges and structures, makes full use of the gradient relationship between the image energy, and adds fuzzy kernel operators to the classical variational denoising model. For the convex functional model, the idea of variable separation is adopted, and the two penalty item and relaxation variables are introduced to decompose the unconstrained optimization problem into a series of subproblems, and combined with the alternating split Bregman technique and frame control of the reconciliation, the function is iterated directly. Meanwhile, the threshold operator and contraction technique are introduced to optimize the subproblems for maintaining important edges and details of medical images and overcoming the computational complexity of traditional methods. The simulation results show that compared with the traditional restoration methods, this proposed method improves the signal-to-noise ratio of the image, significantly reduces mean square error, overcomes the ringing effect, and improves the image visual effect, which proves the validity of the method. The method can be more effective applied to clinical diagnosis and subsequent segmentation for it has a good stability and a fast convergence.
出处
《中国医学物理学杂志》
CSCD
2017年第11期1117-1123,共7页
Chinese Journal of Medical Physics
基金
国家自然科学基金(61401150
61472119
61572173)
河南理工大学博士基金(B2013-039)
河南理工大学创新型科研团队计划(T2014-3)
