CAJ | 학술논문

针对大面积破损区域图像修复中Criminisi算法存在修复质量差和时间复杂度高的缺点，提出一种改进的图像修复算法。改进算法将优先权的计算形式由相乘变为相加，并增加梯度数据项对优先权的计算方式。通过结构信息控制优先权，从而优先修复结构信息。设计根据待修补块中心点的梯度大小，使用全局搜索来寻找匹配块，以提高修复质量和速度。对置信度更新的方式进行修正，引入每次匹配的精度作为惩罚因子的参数，以减小误差向下一次迭代的传播。仿真实验显示，改进算法的修复效率比原算法提升了58％到70％，且修复质量的视觉效果有所提升。
침대대면적파손구역도상수복중Criminisi산법존재수복질량차화시간복잡도고적결점，제출일충개진적도상수복산법。개진산법장우선권적계산형식유상승변위상가，병증가제도수거항대우선권적계산방식。통과결구신식공제우선권，종이우선수복결구신식。설계근거대수보괴중심점적제도대소，사용전국수색래심조필배괴，이제고수복질량화속도。대치신도경신적방식진행수정，인입매차필배적정도작위징벌인자적삼수，이감소오차향하일차질대적전파。방진실험현시，개진산법적수복효솔비원산법제승료58％도70％，차수복질량적시각효과유소제승。
Due to shortcomings of the Criminisi algorithm for inpainting large area damaged image with poor repair quality and high time complexity,an improved image inpainting algorithm is presented in this paper.The improved algorithm changes the form of calculation for priority from multiplication to addition,and adds the gradient data to calculate priority.In this algorithm,the priority is controlled by the structure information,and therefore the priority is also given to in-paint structure information.The gradient magnitude of the center of the block to be inpainted is used to improve the quality and speed of inpaint in order to global search for the matching block. The error propagation to the next iteration is reduced by amending the update mode of confidence and introducing matching accuracy each time as parameters of the penalty factor.Simulation ex-periments show that the improved algorithm can increase inpaint efficiency from 58%to 70%and has better reconstruction quality from the visual effect compared to the original algorithm.