Limits on super-resolution and how to break them

IEEE Transactions on Pattern Analysis and Machine Intelligence - Tập 24 Số 9 - Trang 1167-1183 - 2002
S. Baker1, T. Kanade1
1Robotics Institute, Carnegie Mellon University, Pittsburgh, Pa, USA

Tóm tắt

Nearly all super-resolution algorithms are based on the fundamental constraints that the super-resolution image should generate low resolution input images when appropriately warped and down-sampled to model the image formation process. (These reconstruction constraints are normally combined with some form of smoothness prior to regularize their solution.) We derive a sequence of analytical results which show that the reconstruction constraints provide less and less useful information as the magnification factor increases. We also validate these results empirically and show that, for large enough magnification factors, any smoothness prior leads to overly smooth results with very little high-frequency content. Next, we propose a super-resolution algorithm that uses a different kind of constraint in addition to the reconstruction constraints. The algorithm attempts to recognize local features in the low-resolution images and then enhances their resolution in an appropriate manner. We call such a super-resolution algorithm a hallucination or reconstruction algorithm. We tried our hallucination algorithm on two different data sets, frontal images of faces and printed Roman text. We obtained significantly better results than existing reconstruction-based algorithms, both qualitatively and in terms of RMS pixel error.

Từ khóa

#Image resolution #Image reconstruction #Image generation #Image analysis #Image sequence analysis #Information analysis #Algorithm design and analysis #Image recognition #Reconstruction algorithms

Tài liệu tham khảo

10.1109/83.650118 10.1109/34.790425 elad, 1996, Super-Resolution Reconstruction of Image Sequences Adaptive Filtering Approach 10.1109/83.650116 10.1023/A:1026501619075 10.1109/34.93808 10.1109/83.748893 10.1016/1049-9652(91)90045-L huang, 1984, Multi-Frame Image Restoration and Registration, Advances in Computer Vision and Image Processing, 1, 317 horn, 1996, Robot Vision baker, 1999, Hallucinating Faces baker, 1999, Super-Resolution Optical Flow 10.1109/AFGR.2000.840616 10.1109/CVPR.2000.854852 10.1023/A:1007901712605 barbe, 1980, Charge-Coupled Devices, 10.1007/3-540-09832-1 bascle, 1996, Motion Deblurring and Super-Resolution from an Image Sequence, Proc Fourth European Conf Computer Vision, 573 bergen, 1992, Hierarchical Model-Based Motion Estimation, Proc Second European Conf Computer Vision, 237 10.1109/CVPR.1994.323784 10.1006/jvci.1993.1030 10.1109/CVPR.1992.223272 10.1109/CVPR.1988.196317 10.1109/29.56062 10.1109/ICIP.1994.413336 10.1109/83.242363 10.1016/0030-4018(94)90210-0 nalwa, 1993, A Guided Tour of Computer Vision 10.1016/0167-8655(87)90067-5 10.1109/83.605404 10.1109/CVPR.2000.854918 10.1109/CVPR.1999.784968 pratt, 1991, Digital Image Processing press, 1992, Numerical Recipes in C 10.1109/CVPR.1997.609311 10.1007/BF00055148 shekarforoush, 1999, Conditioning Bounds for Multi-Frame Super-Resolution Algorithms 10.1109/83.503915 10.1109/83.287017 10.1109/34.655647 10.1109/ICCV.1998.710766 10.1109/ACV.1998.732850 10.1016/1049-9652(92)90065-6 10.1109/AFGR.1998.670958 de bonet, 1997, Multiresolution Sampling Procedure for Analysis and Synthesis of Texture Images, Computer Graphics Proc Ann Conf Series (SIGGRAPH '97), 361 10.1109/CVPR.1998.698672 chiang, 1997, Imaging-Consistent Super-Resolution, Proc DARPA Image Understanding Workshop 10.1109/CVPR.1997.609422 10.1109/CVPR.2000.855843 10.1109/TCOM.1983.1095851 10.1364/JOSAA.6.001715 cheeseman, 1994, Super-Resolved Surface Reconstruction from Multiple Images 10.1016/0146-664X(81)90092-7 born, 1965, Principles of Optics