基于生成对抗网络的蓝光激光刀视频实时上色方法  

Real-time colorization method for blue-ray laser scalpel video based on generative adversarial network

作　　者：潘晓英 廉佳 贺琪琪 薛玉锋 王昊 罗斌[5] PAN Xiaoying;LIAN Jia;HEQiqi;XUEYufeng;WANG Hao;LUO Bin(Shaanxi Key Laboratory of Network Data Analysis and Intelligent Processing,Xi'an 710100,China;School of Computer Science&.Technology,Xi'an University of Post&.Telecommunications,Xi'an 710100,China;School of Software,Northwestern Polytechnical University,Xi'an 710072,China;National Engineering Laboratory for Air-Earth-Sea Integration Big Data Application Technology,Xi'an 710129,China;State Key Laboratory of Information Engineering in Surveying,Mapping and Remote Sensing,Wuhan University,Wuhan 430079,China)

机构地区：[1]陕西省网络数据分析与智能处理重点实验室,陕西西安710100 [2]西安邮电大学计算机学院,陕西西安710100 [3]西北工业大学软件学院,陕西西安710072 [4]空天地海一体化大数据应用技术国家工程实验室,陕西西安710129 [5]武汉大学测绘遥感信息工程国家重点实验室,湖北武汉430079

出　　处：《武汉大学学报(工学版)》2025年第1期150-162,共13页Engineering Journal of Wuhan University

基　　金：国家重点研究发展计划项目(编号:2019YFC0121502)。

摘　　要：针对蓝光激光刀手术视频过曝无法看清病灶的问题,提出了一种上色方法,该方法在黑白成像下对蓝光激光刀手术视频进行实时上色处理,以达到病灶清晰、手术视频真实流畅的目的。首先,计算输入图像的三通道平均值,删除完全曝光无细节的图像;其次,利用mask掩膜定位剩余正常图像高光区域,对高光点进行修复;同时,提出一种基于生成对抗网络的实时上色网络(real-time generative adversarial network,RTGAN)对修复后的图像进行实时上色处理;最后,对RTGAN输出的彩色图像合成完整视频流。对2组数据集与5种上色网络进行对比性实验分析,结果表明,所提方法有很好的实时性和上色效果,可以对蓝光激光刀手术视频进行实时上色处理,获得令人满意的视觉质量。In order to solve the problem of overexposure of the blue-ray laser scalpel surgery video that prevent clear visualization of lesions,a coloring method is proposed in this paper.The method performs real-time colorization of the blue-ray laser scalpel surgical video under black and white imaging to make the lesions clear and the surgery video true and smooth.First,the average value of the three channels is calculated for the input image,and the fully exposed images with no detail are deleted.Next,the highlight points are restored by using mask to locate the highlight areas of the remaining normal images.At the same time,a real-time generative adversarial network(RTGAN)is proposed to perform real-time color processing on the restored images.Finally,the color image output from the RTGAN is synthesized into a complete video stream.By comparative experimental analysis of the two datasets and five kinds of coloring networks,the results show that the proposed method has good real-time and coloring effect.The proposed method can colorize the surgical video of blue-ray laser scalpel in real time,and obtain satisfactory visual quality.

关 键 词：蓝光激光刀视频 实时上色 神经网络 RTGAN 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]