waterfall
Image Quilting for Texture Synthesis and Transfer碧昂斯资料
Alexei A.Efros&William T.Freeman
Artificial Intelligence Laboratory
Massachutts Institute of Technology
Cambridge,Massachutts02139
www.ai.mit.edu
废黜
The Problem:A texture synthesis algorithm should be able to take a sample of texture and generate an unlimited amount of image data which,while not exactly like the original,will be perceived by humans to be the same texture. Furthermore,it would be uful to be able transfer the texture from one object to he ability to cut and paste material properties on arbitrary objects).In this work we prent an extremely simple algorithm for texture synthesis.We synthesize new texture by taking patches of existing texture and stitching them together in a consistent way.
editorial
Figure1:Demonstration of quilting for texture synthesis and texture transfer.Using the rice texture(upflowergarden
per left), we can synthesize more such texture(upper right).We can also transfer the rice texture onto another image(lower left)for a strikingly different result.
Previous Work:With Heeger and Bergen[4]the quality of texture synthesis reached a level acceptable for u in computer graphics.Their main insight was to model texture in terms of histograms offilter respons at multiple scales and orientations.But the marginal statistics fail to capture important relationships across scales and ori-entations for structured textures.Several attempts have been made to extend this idea to capture a wider range of textures,including De Bonet[1]who samples from conditional distribution over multiple scales,and later Por-tilla et.al.[5]who match bothfirst and cond order properties of wavelet coefficients.While important from a theoretical point of view,neither method is successful at capturing local detail of many structured textures.
书目推荐
Figure2:Quilting texture.Square blocks from the input texture are patched together to synthesize a new texture sample:(a)blocks are chon randomly(similar to[7,6]),(b)the blocks overlap and each new block is chon so as to“agree”with its neighbors in the region of overlap,(c)to reduce blockiness the boundary between blocks is computed as a minimum cost path through the error surface at the overlap.
考满分网Approach–Image Quilting:Here,we outline our patch-bad texture synthesis procedure,image quilting.To synthesize a new texture image,as afirst step let us simply tile it with blocks taken randomly from S B.The result shown on Figure2(a)already looks somewhat reasonable,but for most structured textures it will be quite obvious that the blocks do not match.
As the next step,let us introduce some overlap in the placement of blocks onto the new image.Now,instead of picking a random block,we arch for a block that by some measure agrees with its neighbors along the region of overlap.Figure2(b)shows an improvement in result,however the edges between the blocks are still quite noticeable.
Finally,we will let the blocks have ragged edges which will allow them to better approximate the features in the texture.Now,before placing a chon block into the texture we will look at the error in t
captain是什么意思
he overlap region between it and the other blocks.Wefind a minimum cost path through that error surface and declare that to be the boundary of the new block.Figure2(c)shows the results of this simple modification.
Impact and Future Work:Despite its simplicity,this method[2]works amazingly well when applied to texture synthesis,producing results that are equal or better than the state-of-the-art[3]but at a fraction of the computational cost.We have also extended our method to the novel task of“texture transfer”with promising results.Fig.1shows the result of transferring the rice texture onto the man’s face.We hope to extend this to higher-level visual objects beyond texture.
Rearch Support:This rearch was conducted at Mitsubishi Electric Rearch Laboratories,where AAE was a summer intern and WTF was a Senior Rearch Scientist.
References:
[1]J.S.De Bonet.Multiresolution sampling procedure for analysis and synthesis of texture images.In SIGGRAPH
97,pages361–368,1997.
atomic[2]A.A.Efros and W.T.Freeman.Image quilting for texture synthesis and transfer.In ACM SIGGRAPH,2001.In
Computer Graphics Proceedings,Annual Conference Series.
[3]A.A.Efros and T.K.Leung.Texture synthesis by non-parametric sampling.In International Conference on
Computer Vision,pages1033–1038,Corfu,Greece,September1999.逡巡
[4]David J.Heeger and James R.Bergen.Pyramid-bad texture analysis/synthesis.In SIGGRAPH95,pages
229–238,1995.
[5]J.Portilla and E.P.Simoncelli.A parametric texture model bad on joint statistics of complex wavelet coeffi-
cients.International Journal of Computer Vision,40(1):49–71,December2000.
[6]Emil Praun,Adam Finkelstein,and Hugues Hoppe.Lapped textures.In SIGGRAPH00,pages465–470,2000.
[7]Y.Xu,B.Guo,and H.-Y.Shum.Chaos mosaic:Fast and memory efficient texture synthesis.Technical Report
MSR-TR-2000-32,Microsoft Rearch,April2000.
