扩展相位拉伸变换的血管造影图像锐化

Angiographic Sharpening with Extended Phase Stretch Transform

针对目前传统血管造影图像锐化增强后大量细小血管变得模糊不清或丢失,甚至增强图像中血管周围产生大量背景噪声,提出一种相位拉伸核函数,形成基于扩展相位拉伸变换的血管造影图像增强算法.该算法将"S"型群延迟相位滤波器推广到线性群延迟相位滤波器,并从理论上证明,这种线性相位拉伸的逆变换相位近似于原图的归一化二阶梯度,将高频特征传统的梯度极值表达转换为角度表达,从而更有利于凸显、增强图像中的高频特征.同时,该算法还结合相对总变分理论,将像素的邻域总变分测度与邻域内在变分测度用于增强过程,使算法更好地突出边缘轮廓与结构纹理,抑制细碎杂乱纹理与背景噪声,克服了目前方法存在的不足.利用Matlab软件平台对DeepLesion, OASIS等数据集中的部分图像数据进行实验,与传统相位拉伸变换增强算法、基于相位一致性的血管造影图像锐化算法等进行对照分析,结果表明,增强后图像上细小血管明显清晰,背景噪声得到有效的抑制,平均梯度和信息熵提高均在50%左右,证明了算法的优越性与实用性.

Facing the problem of a large number of small blood vessels becoming blurred or missing in sharpened angiography image based on conventional enhancement algorithm, and even generating a lot of background noise, a phase stretching kernel function was developed, which resulted in an angiography image enhancement algorithm, called extended phase stretch transform enhancement algorithm. This algorithm expended the ‘S’ type group delay’s phase filter to the linear group delay phase filter, and it was proved theoretically that the inverse transformation’s phase of this phase stretch was approximately the normalized second-order gradient of the original image. The traditional gradient extremum expression of high frequency features was converted into angle expression, which was more conducive to highlight and enhance the high frequency features in images. In addition, combined with the relative total variation theory, the windowed total variation measure and the windowed inherent variation measure of pixels were used in enhance process, which made the algorithm better capable of highlighting the edge contour and texture, suppressing the fine and disorderly texture and background noise, overcoming the shortcomings of current methods. Matlab software platform was used to conduct experiments on some image data in data sets such as DeepLesion and OASIS, compared with the enhancement algorithm based on traditional phase stretch transform and image sharpening algorithm upon phase consistency, the experimental results showed that the small blood vessels in the enhanced image were much clear and the background noise was effectively suppressed. The average gradient and information entropy increased by about 50%, which demonstrated the superiority and practicability of the algorithm.