Super-Resolution Imager via Compressive Sensing
In this paper, we propose a novel imager that can acquire super-resolution (SR) images with significantly fewer sensors. The theoretical basis of this imager is compressive sensing (CS) theory, which calls for a measurement matrix with good properties for effective reconstruction, such as RIP 3. Such a property indicates that entries of the received signal are effectively aliased. In our imager we use an optic effect called spherical aberration to achieve such aliased measurement of light intensity (the signal), thus realizing an ideal measurement matrix. The original image can then be efficiently reconstructed through the Alternating Direction Method (ADM) 2. The implementation of the proposed imager needs only replace an ordinary lens with a spherical lens of large curvature, with almost no additional cost, in contrast with existing complex systems, such as the single pixel camera using the micro-mirror device 12. Simulation results show that despite its simplicity, the performance of the proposed imager is comparable with traditional CS models (most of which are difficult for physical implementation). Further, since the lens is a linear shift-invariant (LSI) system, FFT can be incorporated into the ADM algorithm to accelerate the reconstruction 8, adding to its advantage over some other CS-based imagers.
super-resolution compressive sensing aliased measurement spherical aberration Alternating Direction Method
Qi Wang Guangming Shi
School of Electronic Engineering, Xidian University, Xi’an, China
国际会议
2010 IEEE 10th International Conference on Signal Processing(第十届信号处理国际会议 ICSP 2010)
北京
英文
956-959
2010-08-24（万方平台首次上网日期，不代表论文的发表时间）