CG Term Project Fragment-Based Image Completion
Outline Motivation Method Expected results Current results Reference
Motivation Photo editing (Site Engineering) Using less space to store images Applications to digital communication Photo editing:例如去掉照片裡的路人 Using less space to store large images (texture synthesis):把照片裡重複的部分去掉,以後再弄回來 Applications on digital communication:傳圖片所需的bit rate可以減少(但計算量增加)
Methods Fast approximation search for most similar fragment Image composition
Fast Approximation Gaussian blur
Fast Approximation (Cont.) Gaussian filter
Fast Approximation (Cont.) Down sampling, up sampling, and pyramid
Fast Approximation (Cont.) Expected results: Input, ¯C + α: 100 lines are randomly sampled from a 512 by 512 image and superimposed on a white background. (b-e) YlT(l) for l = 4, . . . , 1. (e) the result of our approximation. (f) Source image.
Search for Most Similar Fragment (a-b) Input color and inverse matte. (c) several matching neighborhoods outlined with circles of the same color, the target center marked by a cross.
Image composition Use α channel Similar to transparency Apply the equations to each level of the pyramid, and then reconstruct the image with it. F: foreground B: background C: Color component Alpha: 透明度 我想reconstruct的細節應該不用講了,主要的精神就是用透明度 (其實是我搞不懂為什麼要reconstruct而不直接用Gaussian)
Expected Results Construct a plausible image Apply to texture synthesis Try to make the search faster - took about 2.5 hours for a 384 x 256 image ( > 0.7 with a 2.4GHz CPU)
Current Results Approximation 左上:original (關山騎腳踏車!!) 右下:approximate結果
Current Results (Cont.)
Reference DRORI, I., COHEN-OR, D., AND YESHURUN, H. 2003. Fragment-Based Image Completion. In Proceedings of ACM SIGGRAPH 2003. GONZALEZ, R. C., AND WOODS, R. E. 2002. Digital Image Processing. Prentice Hall. BURT, P. J., AND ADELSON, E. H. 1985. Merging images through pattern decomposition. Applications of Digital Image Processing VIII 575, 173–181.