CAJ | 학술논문

传统的子空间非纹理图像修复算法只考虑结构信息特征子空间，对小区域破损部分需要通过多维搜索，图像修复效果不好。提出一种基于加权子空间分流拟合的图像全息修复算法，采用阵列流型全息向量对两子空间进行相异性加权，同时考虑结构信息特征和纹理信息特征，并采用最大似然算法对图像区域像素元素实现分流拟合修复，拟合修复过程中避免了对小区域元素多维搜索，实现全息修复。仿真结果表明，采用该算法进行严重受损图像修复，修复效果较好，PSNR值和GPU加速效果都有明显提高，在处理处理破损图像中具有一定的优越性。
전통적자공간비문리도상수복산법지고필결구신식특정자공간，대소구역파손부분수요통과다유수색，도상수복효과불호。제출일충기우가권자공간분류의합적도상전식수복산법，채용진렬류형전식향량대량자공간진행상이성가권，동시고필결구신식특정화문리신식특정，병채용최대사연산법대도상구역상소원소실현분류의합수복，의합수복과정중피면료대소구역원소다유수색，실현전식수복。방진결과표명，채용해산법진행엄중수손도상수복，수복효과교호，PSNR치화GPU가속효과도유명현제고，재처리처리파손도상중구유일정적우월성。
Traditional subspace non-texture image repair algorithm only considers structural information feature, the multi-dimensional search is need in the small area repair, and the image repair effect is bad. An improved image holographic re-pair algorithm is proposed based on weighted subspace split-flow fitting, and the structural information features and texture features are taken into consideration. The maximum likelihood algorithm is used for split-flow fitting, and the holographic repair is realized. Simulation result shows that the severely damaged images can be repaired with this method effectively, the PSNR value and GPU acceleration effect are improved, it has certain superiority in the image processing.