Joint Learning of Measurement Matrix and Signal Reconstruction via Deep Learning

Mdrafi, Robiulhossain; Gürbüz, Ali Cafer

doi:10.1109/tci.2020.2983153

Cited by 21 publications

(12 citation statements)

References 26 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We compare CASNet with ten representative state-of-theart CS networks: ReconNet [23], ISTA-Net + [32], CSNet + [24], DPA-Net [26], ConvMMNet [73], OPINE-Net + [36], AMP-Net [37], SCSNet [40] BCS-Net [39] and COAST [41]. ReconNet, CSNet + , DPA-Net and ConvMMNet are traditional network-based methods; ISTA-Net + , OPINE-Net + and AMP-Net are traditional unfolding methods; SCSNet is scalable with a hierarchical structure; BCS-Net is saliency-based and achieves CS ratio allocation with a multi-channel architecture; COAST achieves scalability by generalizing to arbitrary sampling matrices.…”

Section: Comparison With State-of-the-artsmentioning

confidence: 99%

Content-Aware Scalable Deep Compressed Sensing

Chen

Zhang

2022

IEEE Trans. on Image Process.

View full text Add to dashboard Cite

To more efficiently address image compressed sensing (CS) problems, we present a novel content-aware scalable network dubbed CASNet which collectively achieves adaptive sampling rate allocation, fine granular scalability and high-quality reconstruction. We first adopt a data-driven saliency detector to evaluate the importances of different image regions and propose a saliency-based block ratio aggregation (BRA) strategy for sampling rate allocation. A unified learnable generating matrix is then developed to produce sampling matrix of any CS ratio with an ordered structure. Being equipped with the optimizationinspired recovery subnet guided by saliency information and a multi-block training scheme preventing blocking artifacts, CASNet jointly reconstructs the image blocks sampled at various sampling rates with one single model. To accelerate training convergence and improve network robustness, we propose an SVD-based initialization scheme and a random transformation enhancement (RTE) strategy, which are extensible without introducing extra parameters. All the CASNet components can be combined and learned end-to-end. We further provide a fourstage implementation for evaluation and practical deployments. Experiments demonstrate that CASNet outperforms other CS networks by a large margin, validating the collaboration and mutual supports among its components and strategies. Codes are available at https://github.com/Guaishou74851/CASNet.

show abstract

Section: Comparison With State-of-the-artsmentioning

confidence: 99%

Content-Aware Scalable Deep Compressed Sensing

Chen

Zhang

2022

IEEE Trans. on Image Process.

View full text Add to dashboard Cite

show abstract

“…In [ 17 ], a fully connected layer was added to the ReconNet architecture to learn the measurements from the scene image. The work in [ 20 ] used the method of a joint learning measurement matrix, which greatly improves the reconstruction quality. These methods of learning measurement matrices can obtain matrices that are more suitable for natural images, but the training process for these matrices is complicated and at the same time highly dependent on the training set.…”

Section: Basic Concepts and Related Workmentioning

confidence: 99%

Dual Optical Path Based Adaptive Compressive Sensing Imaging System

Lü

Xue

et al. 2021

Sensors

View full text Add to dashboard Cite

Compressive Sensing (CS) has proved to be an effective theory in the field of image acquisition. However, in order to distinguish the difference between the measurement matrices, the CS imaging system needs to have a higher signal sampling accuracy. At the same time, affected by the noise of the light path and the circuit, the measurements finally obtained are noisy, which directly affects the imaging quality. We propose a dual-optical imaging system that uses the bidirectional reflection characteristics of digital micromirror devices (DMD) to simultaneously acquire CS measurements and images under the same viewing angle. Since deep neural networks have powerful modeling capabilities, we trained the filter network and the reconstruction network separately. The filter network is used to filter the noise in the measurements, and the reconstruction network is used to reconstruct the CS image. Experiments have proved that the method we proposed can filter the noise in the sampling process of the CS system, and can significantly improve the quality of image reconstruction under a variety of algorithms.

show abstract

“…However, this approach can lead to the presence of ringing artifacts and blur in the reconstructed image I Ã due to the absence of high-frequency coefficients. 20,31 Variable density FSI suggests a sampling probability determined by the distance from the center of the Fourier spectrum, as shown in Eq. ( 6).…”

Section: Introductionmentioning

confidence: 99%

“…Consequently, only the low-frequency Fourier coefficients are acquired. However, this approach can lead to the presence of ringing artifacts and blur in the reconstructed image I* due to the absence of high-frequency coefficients 20 , 31 …”

Section: Introductionmentioning

confidence: 99%

Self-adaptive sampling for Fourier single-pixel imaging using probability estimation

Tey,

Tham,

Phua

et al. 2024

J. Electron. Imag.

View full text Add to dashboard Cite

Fourier single-pixel imaging (FSI) produces images through the acquisition of Fourier domain information using a single-pixel detector and a sequence of light modulation patterns. Conventional sampling approach in FSI tends to perform poorly when it comes to capturing the intricate details present in high-frequency components of the image. The variable density sampling method follows a predefined mechanism where the power of image information decreases when frequency increases. To enhance the sampling efficiency, we propose a self-adaptive sampling method to dynamically determine the order of the illumination pattern based on the scene's spectrum distribution through the probability estimation of the low-frequency samples. The image is subsequently reconstructed through compressed sensing technique. Our results indicate improved image quality even at low sampling ratios. Unlike existing adaptive approaches, the proposed method does not require dataset training or redundant sampling. Instead, it exhibits remarkable versatility by adapting to various types of images.

show abstract

Joint Learning of Measurement Matrix and Signal Reconstruction via Deep Learning

Cited by 21 publications

References 26 publications

Content-Aware Scalable Deep Compressed Sensing

Content-Aware Scalable Deep Compressed Sensing

Dual Optical Path Based Adaptive Compressive Sensing Imaging System

Self-adaptive sampling for Fourier single-pixel imaging using probability estimation

Contact Info

Product

Resources

About