Compressive image recovery using recurrent generative model

Dave, Akshat; Kumar, Anil; Vadathya,; Mitra, Kaushik

doi:10.1109/icip.2017.8296572

Cited by 16 publications

(14 citation statements)

References 25 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We can estimate P (x) and sample from the model [21]- [23], or directly generate new samples from P (x) without explicitly estimating the distribution [19], [24]. Dave et al [25] use a spatial long-short-term memory network to learn the distribution P (x); to solve linear inverse problems, they solve a maximum a posteriori estimationmaximizing P (x) for x such that y = Ax. Nguyen et al [26] use a discriminative network and denoising autoencoders to implicitly learn the joint distribution between the image and its label P (x, y), and they generate new samples by sampling the joint distribution P (x, y), i.e., the network, with an approximated Metropolisadjusted Langevin algorithm.…”

Section: Related Workmentioning

confidence: 99%

One Network to Solve Them All — Solving Linear Inverse Problems Using Deep Projection Models

Chang

Póczos

et al. 2017

2017 IEEE International Conference on Computer Vision (ICCV)

272

226

View full text Add to dashboard Cite

While deep learning methods have achieved state-of-the-art performance in many challenging inverse problems like image inpainting and super-resolution, they invariably involve problem-specific training of the networks. Under this approach, different problems require different networks. In scenarios where we need to solve a wide variety of problems, e.g., on a mobile camera, it is inefficient and costly to use these specially-trained networks. On the other hand, traditional methods using signal priors can be used in all linear inverse problems but often have worse performance on challenging tasks. In this work, we provide a middle ground between the two kinds of methods -we propose a general framework to train a single deep neural network that solves arbitrary linear inverse problems. The proposed network acts as a proximal operator for an optimization algorithm and projects non-image signals onto the set of natural images defined by the decision boundary of a classifier. In our experiments, the proposed framework demonstrates superior performance over traditional methods using a wavelet sparsity prior and achieves comparable performance of specially-trained networks on tasks including compressive sensing and pixel-wise inpainting.

show abstract

Section: Related Workmentioning

confidence: 99%

One Network to Solve Them All — Solving Linear Inverse Problems Using Deep Projection Models

Chang

Póczos

et al. 2017

2017 IEEE International Conference on Computer Vision (ICCV)

272

226

View full text Add to dashboard Cite

show abstract

“…SLM may function on a transmission or reflection principle. To enhance the effectiveness (shorten the time) studies are underway on adaptive algorithms that generate pixel masks, which may considerably minimise the number of required measurements [16][17][18]. A solution that still remains universal is the initial concept of applying mask sets, which meet the property of RIP (Restricted Isometry Property).…”

Section: Single Pixel Camera With Compressed Sensingmentioning

confidence: 99%

Simulation of the Operation of a Single Pixel Camera with Compressive Sensing in the Long-Wave Infrared

Szajewska¹

2021

PAR

View full text Add to dashboard Cite

Imaging with the use of a single pixel camera and based on compressed sensing (CS) is a new and promising technology. The use of CS allows reconstruction of images in various spectrum ranges depending on the spectrum sensibility of the used detector. During the study image reconstruction was performed in the LWIR range based on a thermogram from a simulated single pixel camera. For needs of reconstruction CS was used. A case analysis showed that the CS method may be used for construction of infrared-based observation single pixel cameras. This solution may also be applied in measuring cameras. Yet the execution of a measurement of radiation temperature requires calibration of results obtained by CS reconstruction. In the study a calibration method of the infrared observation camera was proposed and studies were carried out of the impact exerted by the number of measurements made on the quality of reconstruction. Reconstructed thermograms were compared with reference images of infrared radiation. It has been ascertained that the reduction of the reconstruction error is not directly in proportion to the number of collected samples being collected. Based on a review of individual cases it has been ascertained that apart from the number of collected samples, an important factor that affects the reconstruction fidelity is the structure of the image as such. It has been proven that estimation of the error for reconstructed thermograms may not be based solely on the quantity of executed measurements.

show abstract

“…A class of deep learning based solution involves learning of regularizers or proximal mapping stage and then iteratively solving a MAP problem. Methods like [21], [22], [23] fall under this category. Another class of algorithm is designed as a feed-forward deep neural network that has either been trained in a supervised or self-supervised manner.…”

Section: Image Reconstructionmentioning

confidence: 99%

FlatNet: Towards Photorealistic Scene Reconstruction from Lensless Measurements

Khan

Sundar

Boominathan

et al. 2020

IEEE Trans. Pattern Anal. Mach. Intell.

Self Cite

View full text Add to dashboard Cite

Lensless imaging has emerged as a potential solution towards realizing ultra-miniature cameras by eschewing the bulky lens in a traditional camera. Without a focusing lens, the lensless cameras rely on computational algorithms to recover the scenes from multiplexed measurements. However, the current iterative-optimization-based reconstruction algorithms produce noisier and perceptually poorer images. In this work, we propose a non-iterative deep learning-based reconstruction approach that results in orders of magnitude improvement in image quality for lensless reconstructions. Our approach, called FlatNet, lays down a framework for reconstructing high-quality photorealistic images from mask-based lensless cameras, where the camera's forward model formulation is known. FlatNet consists of two stages: (1) an inversion stage that maps the measurement into a space of intermediate reconstruction by learning parameters within the forward model formulation, and (2) a perceptual enhancement stage that improves the perceptual quality of this intermediate reconstruction. These stages are trained together in an end-to-end manner. We show high-quality reconstructions by performing extensive experiments on real and challenging scenes using two different types of lensless prototypes: one which uses a separable forward model and another, which uses a more general non-separable cropped-convolution model. Our end-to-end approach is fast, produces photorealistic reconstructions, and is easy to adopt for other mask-based lensless cameras.

show abstract

Compressive image recovery using recurrent generative model

Cited by 16 publications

References 25 publications

One Network to Solve Them All — Solving Linear Inverse Problems Using Deep Projection Models

One Network to Solve Them All — Solving Linear Inverse Problems Using Deep Projection Models

Simulation of the Operation of a Single Pixel Camera with Compressive Sensing in the Long-Wave Infrared

FlatNet: Towards Photorealistic Scene Reconstruction from Lensless Measurements

Contact Info

Product

Resources

About