Motion Deblurring: Algorithms and Systems

Half Title
Title Page
Imprints
Contents
List of contributors
Preface
1 Mathematical models and practical solvers for uniform motion deblurring
	1.1 Non-blind deconvolution
		1.1.1 Regularized approaches
		1.1.2 Iterative approaches
		1.1.3 Recent advancements
		1.1.4 Variable splitting solver
		1.1.5 A few results
	1.2 Blind deconvolution
		1.2.1 Maximum marginal probability estimation
		1.2.2 Alternating energy minimization
		1.2.3 Implicit edge recovery
		1.2.4 Explicit edge prediction for very large PSF estimation
		1.2.5 Results and running time
2 Spatially varying image deblurring
	2.1 Review of image deblurring methods
	2.2 A unified camera shake blur model
		2.2.1 Blur matrices
		2.2.2 Spatially-varying deconvolution
	2.3 Single image deblurring using motion density functions
		2.3.1 Optimization formulation
		2.3.2 Our system for MDF-based deblurring
		2.3.3 Experiments and results
	2.4 Image deblurring using inertial measurement sensors
		2.4.1 Deblurring using inertial sensors
		2.4.2 Deblurring system
		2.4.3 Results
	2.5 Generating sharp panoramas from motion-blurred videos
		2.5.1 Motion and duty cycle estimation
		2.5.2 Experiments
		2.5.3 Real videos
	2.6 Discussion
3 Hybrid-imaging for motion deblurring
	3.1 Introduction
	3.2 Fundamental resolution tradeoff
	3.3 Hybrid-imaging systems
	3.4 Shift invariant PSF image deblurring
		3.4.1 Parametric motion computation
		3.4.2 Shift invariant PSF estimation
		3.4.3 Image deconvolution
		3.4.4 Examples – shift invariant PSF
		3.4.5 Shift-invariant PSF optimization
		3.4.6 Examples – optimized shift invariant PSF
	3.5 Spatially-varying PSF image deblurring
		3.5.1 Examples – spatially varying PSF
	3.6 Moving object deblurring
	3.7 Discussion and summary
4 Efficient, blind, spatially-variant deblurring for shaken images
	4.1 Introduction
	4.2 Modelling spatially-variant camera shake blur
		4.2.1 Components of camera motion
		4.2.2 Motion blur and homographies
		4.2.3 Camera calibration
		4.2.4 Uniform blur as a special case
	4.3 The computational model
	4.4 Blind estimation of blur from a single image
		4.4.1 Updating the blur kernel
		4.4.2 Updating the latent image
	4.5 Efficient computation of the spatially-variant model
		4.5.1 A locally-uniform approximation for camera shake
		4.5.2 Updating the blur kernel
		4.5.3 Updating the latent image fast, non-iterative non-blind deconvolution
	4.6 Single-image deblurring results
		4.6.1 Limitations and failures
	4.7 Implementation
	4.8 Conclusion
5 Removing camera shake in smartphones without hardware stabilization
	5.1 Introduction
	5.2 Image acquisition model
		5.2.1 Space-invariant model
		5.2.2 Space-variant model
	5.3 Inverse problem
		5.3.1 MAP and beyond
		5.3.2 Getting more prior information
		5.3.3 Patch based
	5.4 Pinhole camera model
	5.5 Smartphone application
		5.5.1 Space-invariant implementation
		5.5.2 Space-variant implementation
	5.6 Evaluation
	5.7 Conclusions
6 Multi-sensor fusion for motion deblurring
	6.1 Introduction
	6.2 Hybrid-speed sensor
	6.3 Motion deblurring
		6.3.1 Motion flow estimation
		6.3.2 Motion warping
		6.3.3 PSF estimation and motion deblurring
	6.4 Depth map super-resolution
		6.4.1 Initial depth estimation
		6.4.2 Joint bilateral upsampling
		6.4.3 Results and discussion
	6.5 Extensions to low-light imaging
		6.5.1 Sensor construction
		6.5.2 Processing pipeline
		6.5.3 Preliminary results
	6.6 Discussion and summary
7 Motion deblurring using fluttered shutter
	7.1 Related work
	7.2 Coded exposure photography
	7.3 Image deconvolution
		7.3.1 Motion model
	7.4 Code selection
	7.5 Linear solution for deblurring
		7.5.1 Background estimation
		7.5.2 Motion generalization
	7.6 Resolution enhancement
	7.7 Optimized codes for PSF estimation
		7.7.1 Blur estimation using alpha matting
		7.7.2 Motion from blur
		7.7.3 Code selection
		7.7.4 Results
	7.8 Implementation
	7.9 Analysis
		7.9.1 Noise analysis
		7.9.2 Resolution analysis
	7.10 Summary
8 Richardson–Lucy deblurring for scenes under a projective motion path
	8.1 Introduction
	8.2 Related work
	8.3 The projective motion blur model
	8.4 Projective motion Richardson–Lucy
		8.4.1 Richardson–Lucy deconvolution algorithm
		8.4.2 Projective motion Richardson–Lucy algorithm
		8.4.3 Gaussian noise
	8.5 Motion estimation
	8.6 Experiment results
		8.6.1 Convergence analysis
		8.6.2 Noise analysis
		8.6.3 Qualitative and quantitative analysis
		8.6.4 Comparisons with spatially invariant method
		8.6.5 Real examples
	8.7 Discussion and conclusion
		8.7.1 Conventional PSF representation versus projective motion blur model
		8.7.2 Limitations
		8.7.3 Running time analysis
9 HDR imaging in the presence of motion blur
	9.1 Introduction
	9.2 Existing approaches to HDRI
		9.2.1 Spatially-varying pixel exposures
		9.2.2 Multiple exposures (irradiance)
		9.2.3 Multiple exposures (direct)
	9.3 CRF, irradiance estimation and tone-mapping
		9.3.1 Estimation of inverse CRF and irradiance
		9.3.2 Tone-mapping
	9.4 HDR imaging under uniform blurring
	9.5 HDRI for non-uniform blurring
		9.5.1 Image accumulation
		9.5.2 TSF and its estimation
		9.5.3 PSF estimation
		9.5.4 Irradiance image recovery
	9.6 Experimental results
	9.7 Conclusions and discussions
10 Compressive video sensing to tackle motion blur
	10.1 Introduction
		10.1.1 Video compressive sensing to handle complex motion
	10.2 Related work
	10.3 Imaging architecture
		10.3.1 Programmable pixel compressive camera
		10.3.2 Prototype P2C2
		10.3.3 P2C2 as a underdetermined linear system
	10.4 High-speed video recovery
		10.4.1 Transform domain sparsity
		10.4.2 Brightness constancy as temporal redundancy
	10.5 Experimental results
		10.5.1 Simulation on high-speed videos
		10.5.2 Results on P2C2 prototype datasets
	10.6 Conclusions
11 Coded exposure motion deblurring for recognition
	11.1 Motion sensitivity of iris recognition
	11.2 Coded exposure
		11.2.1 Sequence selection for image capture
		11.2.2 Blur estimation
		11.2.3 Deblurring
	11.3 Coded exposure performance on iris recognition
		11.3.1 Synthetic experiments
		11.3.2 Real image experiments
	11.4 Barcodes
	11.5 More general subject motion
	11.6 Implications of computational imaging for recognition
	11.7 Conclusion
12 Direct recognition of motion-blurred faces
	12.1 Introduction
		12.1.1 Related work
	12.2 The set of all motion-blurred images
		12.2.1 Convolution model for blur
		12.2.2 The set of all blurred images versus the set of motion-blurred images
	12.3 Bank of classifiers approach for recognizing motion-blurred faces
	12.4 Experimental evaluation
		12.4.1 Sensitivity analysis of the BoC approach
		12.4.2 Performance evaluation on synthetically generated motion-blurred images
		12.4.3 Performance evaluation on the real dataset REMOTE
	12.5 Discussion
13 Performance limits for motion deblurring cameras
	13.1 Introduction
	13.2 Performance bounds for flutter shutter cameras
		13.2.1 Optimal flutter shutter performance
	13.3 Performance bound for motion-invariant cameras
		13.3.1 Space–time analysis
		13.3.2 Optimal motion-invariant performance
	13.4 Simulations to verify performance bounds
	13.5 Role of image priors
	13.6 When to use computational imaging
		13.6.1 Rule of thumb
	13.7 Relationship to other computational imaging systems
		13.7.1 Example computational cameras
	13.8 Summary and discussion
Index