Visual Comput (2007)23:833-841 D0110.1007/s00371-007-0137-4 ORIGINAL ARTICLE Chunxiao Liu Image completion based on views of large Yanwen Guo Liang Pan displacement Qunsheng Peng Fuyan Zhang Published online:19 June 2007 Abstract This paper presents an transformed PSRs,we develop a new Springer-Verlag 2007 algorithm for image completion image repairing algorithm,coupled based on the views of large displace- with graph cut based image stitch- ment.A distinct from most existing ing,texture synthesis based image image completion methods,which inpainting,and image fusion based exploit only the target image's own hole filling,to complete the missing information to complete the damaged regions seamlessly.Finally,the ghost regions,our algorithm makes full effect between the repaired region use of a large displacement view and its surroundings is eliminated (LDV)of the same scene,which by Poisson image blending.Our introduces enough information to algorithm effectively preserves the C.Liu(✉).Y.Guo-L.Pan·Q.Peng resolve the original ill-posed prob- structure information on the missing State Key Lab of CAD&CG. lem.To eliminate any perspective area of the target image and produces Zhejiang University,310027,Hangzhou, China distortion during the warping of the a repaired result comparable to its liuchunxiao.ywguo.panliang.peng} LDV image,we first decompose original appearance.Experiments @cad.zju.edu.cn the target image and the LDV one show the effectiveness of our method. Y.Guo.F.Zhang into several corresponding planar National Laboratory for Novel Software scene regions (PSRs)and transform Keywords Image completion. Technology,Nanjing University, the candidate PSRs on the LDV Large displacement view.Image 210093,Nanjing.China image onto their correspondences stitching.Texture synthesis ywguo,fyzhang@graphics.nju.edu.cn on the target image.Then using the 1 Introduction may run into the view of our camera.The problem is,there- fore,how to generate a visually pleasing result after removal Image completion is the task of filling in the missing or of the intruder from the image.It would be difficult to com- damaged regions of an image with the available informa- plete the target image by itself,especially for missing re- tion from the same or another image to generate a per- gions with complex structures.On the other hand,if we have ceptually satisfactory result.Due to the wide applications another view of the same scene with a large displacement in photo editing,rig removal and special effects produc- that reveals the previously occluded regions,we can faith- tion [6],image completion has received great attention in fully restore the occluded regions on the target image the past decade.Many methods have been brought for- Image completion based on views of large displace- ward so far [22].but most of them are mainly focused on ment is different from traditional image completion.Tra- single image based techniques. ditional methods only make use of the surrounding known Few methods address image completion based on views pixels on the target image to solve its unknown regions. of large displacement.In fact,this provides a feasible way which is in nature an ill-posed problem [21].However, for image completion in our lives.For instance,when we video completion bears a certain similarity to our problem take photos at some famous resorts,an undesired person in adopting other frames to repair the current frame,butVisual Comput (2007) 23: 833–841 DOI 10.1007/s00371-007-0137-4 ORIGINAL ARTICLE Chunxiao Liu Yanwen Guo Liang Pan Qunsheng Peng Fuyan Zhang Image completion based on views of large displacement Published online: 19 June 2007 © Springer-Verlag 2007 C. Liu (✉)· Y. Guo · L. Pan · Q. Peng State Key Lab of CAD&CG, Zhejiang University, 310027, Hangzhou, China {liuchunxiao,ywguo,panliang,peng} @cad.zju.edu.cn Y. Guo · F. Zhang National Laboratory for Novel Software Technology, Nanjing University, 210093, Nanjing, China {ywguo,fyzhang}@graphics.nju.edu.cn Abstract This paper presents an algorithm for image completion based on the views of large displace￾ment. A distinct from most existing image completion methods, which exploit only the target image’s own information to complete the damaged regions, our algorithm makes full use of a large displacement view (LDV) of the same scene, which introduces enough information to resolve the original ill-posed prob￾lem. To eliminate any perspective distortion during the warping of the LDV image, we first decompose the target image and the LDV one into several corresponding planar scene regions (PSRs) and transform the candidate PSRs on the LDV image onto their correspondences on the target image. Then using the transformed PSRs, we develop a new image repairing algorithm, coupled with graph cut based image stitch￾ing, texture synthesis based image inpainting, and image fusion based hole filling, to complete the missing regions seamlessly. Finally, the ghost effect between the repaired region and its surroundings is eliminated by Poisson image blending. Our algorithm effectively preserves the structure information on the missing area of the target image and produces a repaired result comparable to its original appearance. Experiments show the effectiveness of our method. Keywords Image completion · Large displacement view · Image stitching · Texture synthesis 1 Introduction Image completion is the task of filling in the missing or damaged regions of an image with the available informa￾tion from the same or another image to generate a per￾ceptually satisfactory result. Due to the wide applications in photo editing, rig removal and special effects produc￾tion [6], image completion has received great attention in the past decade. Many methods have been brought for￾ward so far [22], but most of them are mainly focused on single image based techniques. Few methods address image completion based on views of large displacement. In fact, this provides a feasible way for image completion in our lives. For instance, when we take photos at some famous resorts, an undesired person may run into the view of our camera. The problem is, there￾fore, how to generate a visually pleasing result after removal of the intruder from the image. It would be difficult to com￾plete the target image by itself, especially for missing re￾gions with complex structures. On the other hand, ifwe have another view of the same scene with a large displacement that reveals the previously occluded regions, we can faith￾fully restore the occluded regions on the target image. Image completion based on views of large displace￾ment is different from traditional image completion. Tra￾ditional methods only make use of the surrounding known pixels on the target image to solve its unknown regions, which is in nature an ill-posed problem [21]. However, video completion bears a certain similarity to our problem in adopting other frames to repair the current frame, but