基于双次梯度联合改进Criminisi算法去除超声图像中的人工标记并修复图像的可行性

Feasibility of removing manual marks on ultrasonic image and repairing images based on double gradient combined with improved Criminisi algorithm

目的观察基于双次梯度联合改进Criminisi算法去除超声图像中人工标记并修复图像的可行性。方法选取30幅二维声像图,图中均包含十字、箭头和/或文字标记,以20幅无标记图像作为参考。算法由标记提取模块及图像恢复模块两个部分组成,前者采用双次梯度最大连通面积方法,后者采用改进重加权Criminisi算法;以峰值信噪比(PSNR)和结构相似性(SSIM)作为指标评价修复图像的质量。结果基于双次梯度联合改进Criminisi算法可准确检出超声图像中的人工标记并生成掩模,用于去除标记、恢复图像。相比无标记图像,30幅超声图像提取的标记掩模的平均检测精度和平均错误发现率分别为0.96和0.63,修复图像的平均PSNR及SSIM分别为46.78 dB和0.99。结论基于双次梯度联合改进Criminisi算法可有效去除超声图像中的人工标记并修复图像。

Objective To observe the feasibility of removing artificial labels on ultrasonic images and repairing images based on double gradient combined with improved Criminisi algorithm.Methods Totally 30 two-dimensional ultrasound images containing markers of cross,arrow and text were selected,while 20 unmarked images were taken as references.The algorithm consisted of two parts,i.e.mark extraction module and image restoration module,the former was established using double gradient maximum connected area which could completely extract the label region and screen out the normal region,and the latter was designed with improved reweighted Criminisi algorithm for image restoration.The peak signal to noise ratio(PSNR)and structural similarity(SSIM)were taken as quality evaluation indexes.Results Compared with original unmarked images,Criminisi algorithm based on double gradient could accurately detect artificial marks on ultrasound images,remove various labels and repair images.The average detection accuracy and average false discovery rate of 30 ultrasound images was 0.96 and 0.63,respectively.The average PSNR and SSIM of images recovered by Criminisi algorithm was 46.78 dB and 0.99,respectively.Conclusion The artificial labels on ultrasonic images could be effectively removed and the images could be repaired based on double gradient combined with improved Criminisi algorithm.