基于CMOS图像传感器的压缩感知成像系统研究
|
摘要
近年来提出的压缩感知理论与传统奈奎斯特采样定理不同,它是一种能够将信号的采样和压缩同时进行的新理论。压缩感知理论指出,只要信号在某个变换域上是稀疏的,那么就可以用一个与变换基不相关的测量矩阵将变换所得到的高维信号投影到一个低维空间上,然后通过求解一个最优化问题就可以从这些少量的投影中以高概率重构出原信号。
传统的CMOS图像传感器通过模拟信号读出电路读取像素阵列,然后将其转换为数字信号进行奈奎斯特采样,再丢弃大量冗余数据进行压缩编码。将压缩感知技术引入CMOS图像传感器中,可以在图像光电转换之后,直接对模拟信号进行压缩采样,将信号采集和压缩同时进行,再经过A/D转换,这样就可以有效减少输入A/D转换模块的数据量,提高A/D转换的速度,降低系统功耗。
鉴于现有的基于CMOS图像传感器的压缩感知成像算法对原有的CMOS图像传感器电路结构改动较大,附加电路较多不利于硬件实现的缺点,本文提出采用并行处理策略对CMOS图像传感器A/D转换前的模拟像素矩阵进行压缩采样,减轻了A/D转换模块的负担,可以在很大程度上降低CMOS图像传感器的功耗,并且该算法实现电路简单,不需要较多的额外附加电路。仿真结果表明,本文所提结构能快速有效的进行测量值的获取,并且重构图像的主客观质量没有明显损失。
测量矩阵是成像系统前端重要的组成部分,Toeplitz结构测量矩阵易于生成和存储,本文将Toeplitz结构测量矩阵应用在成像系统中可以降低系统的复杂度和功耗。重构算法是成像系统后端最主要的研究问题,本文在分析已有算法的基础上并结合所提出的成像结构特点对重构算法也进行了改进。
The theory of compressive sensing proposed in recent years is different from the traditional sampling theory-Nyquist sampling theorem. Compressive sensing theory indicates that if the signal is sparse in a transform domain, then we can make a transformed high-dimensional signal to project onto a low-dimensional space using a measurement matrix which is not related to the transform base. The original signal can be reconstructed with high probability from these small measurements by solving a optimization problem
Traditional CMOS image sensor obtains the pixel array by analog readout circuit, and then converts them to digital signals before Nyquist-sampling. A lot of redundant data is discarded before compression. Taking the compressive sensing technology broach to the CMOS image sensors, it can use compressive sensing on the analog signal directly after the photoelectric conversion. Compressive sensing can acquire and compress the signal at the same time, so the data passing through the A/D converter can be reduced. This can improve the speed of A/D conversion and cut down power consumption.
In view of the existing CMOS image sensor systems which based on image compressive sensing algorithm change the structure of the sensor circuit a lot, and are difficult implemented in hardware because of too much additional circuit. This thesis presents a parallel processing strategy for analog pixel matrix of CMOS image sensor. It can reduce the burden of A/D converter module, also can reduce the power consumption of CMOS image sensors. The implementation of the circuit is simple, and it does not need more additional circuit. Simulation results show that the proposed structure can get the measurements quickly and effectively. The subjective and objective quality of the reconstructed image has little loss.
Measurement matrix is an important part of the front-end of the imaging system, Toeplitz-structure measurement matrix generates and stores very easily. Applications in the imaging system can reduce the complexity and power consumption of the image system. Reconstruction algorithms is the main object of study at the back-end of the imaging system. Based on analysis of the existing algorithms and considering the structure of the parallel processing architecture, a improved algorithm has been put forward.
传统的CMOS图像传感器通过模拟信号读出电路读取像素阵列,然后将其转换为数字信号进行奈奎斯特采样,再丢弃大量冗余数据进行压缩编码。将压缩感知技术引入CMOS图像传感器中,可以在图像光电转换之后,直接对模拟信号进行压缩采样,将信号采集和压缩同时进行,再经过A/D转换,这样就可以有效减少输入A/D转换模块的数据量,提高A/D转换的速度,降低系统功耗。
鉴于现有的基于CMOS图像传感器的压缩感知成像算法对原有的CMOS图像传感器电路结构改动较大,附加电路较多不利于硬件实现的缺点,本文提出采用并行处理策略对CMOS图像传感器A/D转换前的模拟像素矩阵进行压缩采样,减轻了A/D转换模块的负担,可以在很大程度上降低CMOS图像传感器的功耗,并且该算法实现电路简单,不需要较多的额外附加电路。仿真结果表明,本文所提结构能快速有效的进行测量值的获取,并且重构图像的主客观质量没有明显损失。
测量矩阵是成像系统前端重要的组成部分,Toeplitz结构测量矩阵易于生成和存储,本文将Toeplitz结构测量矩阵应用在成像系统中可以降低系统的复杂度和功耗。重构算法是成像系统后端最主要的研究问题,本文在分析已有算法的基础上并结合所提出的成像结构特点对重构算法也进行了改进。
The theory of compressive sensing proposed in recent years is different from the traditional sampling theory-Nyquist sampling theorem. Compressive sensing theory indicates that if the signal is sparse in a transform domain, then we can make a transformed high-dimensional signal to project onto a low-dimensional space using a measurement matrix which is not related to the transform base. The original signal can be reconstructed with high probability from these small measurements by solving a optimization problem
Traditional CMOS image sensor obtains the pixel array by analog readout circuit, and then converts them to digital signals before Nyquist-sampling. A lot of redundant data is discarded before compression. Taking the compressive sensing technology broach to the CMOS image sensors, it can use compressive sensing on the analog signal directly after the photoelectric conversion. Compressive sensing can acquire and compress the signal at the same time, so the data passing through the A/D converter can be reduced. This can improve the speed of A/D conversion and cut down power consumption.
In view of the existing CMOS image sensor systems which based on image compressive sensing algorithm change the structure of the sensor circuit a lot, and are difficult implemented in hardware because of too much additional circuit. This thesis presents a parallel processing strategy for analog pixel matrix of CMOS image sensor. It can reduce the burden of A/D converter module, also can reduce the power consumption of CMOS image sensors. The implementation of the circuit is simple, and it does not need more additional circuit. Simulation results show that the proposed structure can get the measurements quickly and effectively. The subjective and objective quality of the reconstructed image has little loss.
Measurement matrix is an important part of the front-end of the imaging system, Toeplitz-structure measurement matrix generates and stores very easily. Applications in the imaging system can reduce the complexity and power consumption of the image system. Reconstruction algorithms is the main object of study at the back-end of the imaging system. Based on analysis of the existing algorithms and considering the structure of the parallel processing architecture, a improved algorithm has been put forward.
引文
[1]沈保锁,侯春萍,现代通讯原理,北京：国防工业出版社,2006.1
[2]Duarte M F, Davenport M A, Takbar D, et al. Single-pixel imaging via compressive sampling [J]. IEEE Signal Processing Magazine in Signal Processing Magazine,2008,25(2):83-91.
[3]Robucci R, Chiu L K, Gray J, et al. Compressive sensing on a CMOS separable transform image sensor[C].
[4]Fergus R, Torralba A and Freeman W T. Random lens imaging. MIT Computer Science and Artificial Intelligence Laboratory (MIT CSAIL), Cambridge, MA, Technical Report, Sep. 2006.
[5]Jacques L, Vander gheynst P, Bibet A, et al. CMOS compressed imaging by random convolution[C].
[6]David L. Donoho, Compressed sensing [J]. IEEE Transactions on Information Theory,2006, 52(4):1289-1306
[7]王建英.信号与图像的稀疏分解及初步应用[M].四川成都：西南交通大学出版社,2006
[8]Candes E J, Romberg J and Tao T. Stable signal recovery from incomplete and inaccurate measurements [J].
[9]Candes E, Romberg J, Tao T. Robust uncertainty principles:Exact signal reconstruction from highly incomplete frequency information [J]. IEEE Trans. Information Theory,2006,52 (2):489-509.
[10]Hong Fang, Quanbing Zhang, Sui Wei, Method of image reconstruction based on very sparse random projection, Computer Engineering and Applications,2007,43(22),25-27
[11]E. Candes, T. Tao, Near optimal signal recovery from random projections:Universal encoding strategies IEEE Transactions on Information Theory,2006,52(12),5406-5425
[12]Y Tsaig, D. Donoho, Extensions of compressed sensing, Signal processing,2006,86(3) 549-571
[13]Holger Rauhut, Circulant and Toeplitz matrices in compressed sensing, In proeessing SAPARS'09, Saint Malo,2009
[14]W. Bajwa, J. Haupt, G Raz, S. Wright,Toeplitz-structured compressed sensing matrices. Proc.of Stat. Sig. Proc. Workshop,2007.
[15]Radu Berinde, piotr Indyk, Sparse recovery using sparse random matrices,2008
[16]T. T. Do, T. D. Trany, L.Gan, fast compressive sampling with structurally random matrices, Proceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing, Washington D.C., USA:IEEE,2008,3369-3372
[17]Justin Romberg, compressive sensing by random convolution, SIAM Journal on Imaging Sciences, Nov.2009,2(4),1098-1128
[18]GDavis, Adaptive nonlinear approximations, Ph.D. dissertation, New York, New York university,1994
[19]SIAMJ.Comput, Sparse approximate solutions to linear systems, SIAM Journal on Computing.Apr.1995,24(2),227-234
[20]Scott Shaobing Chen, David L.Donoho, Mieh ael A.Saunders, Atomic decomposition by basis pursuit, society for Industrial and Applied Mathematics,2001,43(1),129-159
[21]David L. Donoho, M. Elad, V Temlyakov, Stable recovery of sparse over complete representations in the presence of noise,IEEE Transactions on Information Theory,2006, 52(1),6-18
[22]M. Lobo, L. Vandenberghe, S. Boyd, H. Lebret, Applications of second-order cone programming, Linear Algebra and its Applications,1998,284(1-3),193-228
[23]石光明,刘丹华,高大化,等.压缩感知理论及其研究进展[J].电子学报,2009,5(37)：1070-1081.
[24]李树涛,魏丹.压缩传感综述[J].自动化学报,2009,35(11)：1369-1377.
[25]L. Gan, Block compressed sensing of natural images, in Proceedings of the International Conference on Digital Signal Processing, Cardiff, UK, July 2007, pp.403-406.
[26]RF Marcia, R M. Willett. Compressive coded aperture superresolution image reconstruction [J]. IEEE Speech and Signal, March 2008,833-836.
[27]Li Chengbo. An Efficient Algorithm for total variation regularization with applications to the single pixel camera and compressive sensing [D]. Houston:Department of Computational and Applied Mathematics, Rice University,2009.
[28]侯金曼,何宁.基于压缩感知的图像快速重建方法[J].计算机工程,2011,37(19)：215-217
[29]S. Mallat, Z, Zhang, Matching pursuits with time-frequency dictionaries, IEEE Transactions on Signal processing,1993,41(12),3397-3415
[30]Joel Tropp, Anna Gilbert, Signal recovery from random measurements via orthogonal matching pursuit, IEEE Transactions on Information Theory,2008,53(12),4655-4666
[31]Pitsianis N P, Brady D J, PortnoyA, et al. Compressive imaging sensors[C].
[32]Li Binqiao, Sun Zhongyan, Xu Jiangtao.Wide dynamic range CMOS image sensor with in-pixel double-exposure and synthesis [J].
[2]Duarte M F, Davenport M A, Takbar D, et al. Single-pixel imaging via compressive sampling [J]. IEEE Signal Processing Magazine in Signal Processing Magazine,2008,25(2):83-91.
[3]Robucci R, Chiu L K, Gray J, et al. Compressive sensing on a CMOS separable transform image sensor[C].
[4]Fergus R, Torralba A and Freeman W T. Random lens imaging. MIT Computer Science and Artificial Intelligence Laboratory (MIT CSAIL), Cambridge, MA, Technical Report, Sep. 2006.
[5]Jacques L, Vander gheynst P, Bibet A, et al. CMOS compressed imaging by random convolution[C].
[6]David L. Donoho, Compressed sensing [J]. IEEE Transactions on Information Theory,2006, 52(4):1289-1306
[7]王建英.信号与图像的稀疏分解及初步应用[M].四川成都：西南交通大学出版社,2006
[8]Candes E J, Romberg J and Tao T. Stable signal recovery from incomplete and inaccurate measurements [J].
[9]Candes E, Romberg J, Tao T. Robust uncertainty principles:Exact signal reconstruction from highly incomplete frequency information [J]. IEEE Trans. Information Theory,2006,52 (2):489-509.
[10]Hong Fang, Quanbing Zhang, Sui Wei, Method of image reconstruction based on very sparse random projection, Computer Engineering and Applications,2007,43(22),25-27
[11]E. Candes, T. Tao, Near optimal signal recovery from random projections:Universal encoding strategies IEEE Transactions on Information Theory,2006,52(12),5406-5425
[12]Y Tsaig, D. Donoho, Extensions of compressed sensing, Signal processing,2006,86(3) 549-571
[13]Holger Rauhut, Circulant and Toeplitz matrices in compressed sensing, In proeessing SAPARS'09, Saint Malo,2009
[14]W. Bajwa, J. Haupt, G Raz, S. Wright,Toeplitz-structured compressed sensing matrices. Proc.of Stat. Sig. Proc. Workshop,2007.
[15]Radu Berinde, piotr Indyk, Sparse recovery using sparse random matrices,2008
[16]T. T. Do, T. D. Trany, L.Gan, fast compressive sampling with structurally random matrices, Proceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing, Washington D.C., USA:IEEE,2008,3369-3372
[17]Justin Romberg, compressive sensing by random convolution, SIAM Journal on Imaging Sciences, Nov.2009,2(4),1098-1128
[18]GDavis, Adaptive nonlinear approximations, Ph.D. dissertation, New York, New York university,1994
[19]SIAMJ.Comput, Sparse approximate solutions to linear systems, SIAM Journal on Computing.Apr.1995,24(2),227-234
[20]Scott Shaobing Chen, David L.Donoho, Mieh ael A.Saunders, Atomic decomposition by basis pursuit, society for Industrial and Applied Mathematics,2001,43(1),129-159
[21]David L. Donoho, M. Elad, V Temlyakov, Stable recovery of sparse over complete representations in the presence of noise,IEEE Transactions on Information Theory,2006, 52(1),6-18
[22]M. Lobo, L. Vandenberghe, S. Boyd, H. Lebret, Applications of second-order cone programming, Linear Algebra and its Applications,1998,284(1-3),193-228
[23]石光明,刘丹华,高大化,等.压缩感知理论及其研究进展[J].电子学报,2009,5(37)：1070-1081.
[24]李树涛,魏丹.压缩传感综述[J].自动化学报,2009,35(11)：1369-1377.
[25]L. Gan, Block compressed sensing of natural images, in Proceedings of the International Conference on Digital Signal Processing, Cardiff, UK, July 2007, pp.403-406.
[26]RF Marcia, R M. Willett. Compressive coded aperture superresolution image reconstruction [J]. IEEE Speech and Signal, March 2008,833-836.
[27]Li Chengbo. An Efficient Algorithm for total variation regularization with applications to the single pixel camera and compressive sensing [D]. Houston:Department of Computational and Applied Mathematics, Rice University,2009.
[28]侯金曼,何宁.基于压缩感知的图像快速重建方法[J].计算机工程,2011,37(19)：215-217
[29]S. Mallat, Z, Zhang, Matching pursuits with time-frequency dictionaries, IEEE Transactions on Signal processing,1993,41(12),3397-3415
[30]Joel Tropp, Anna Gilbert, Signal recovery from random measurements via orthogonal matching pursuit, IEEE Transactions on Information Theory,2008,53(12),4655-4666
[31]Pitsianis N P, Brady D J, PortnoyA, et al. Compressive imaging sensors[C].
[32]Li Binqiao, Sun Zhongyan, Xu Jiangtao.Wide dynamic range CMOS image sensor with in-pixel double-exposure and synthesis [J].