Abstract:The emerging technology of snapshot compressive imaging (SCI) enables capturing high dimensional (HD) data in an efficient way. It is generally implemented by two components: an optical encoder that compresses HD signals into a 2D measurement and an algorithm decoder that retrieves the HD data upon the hardware-encoded measurement. Over a broad range of SCI applications, hyperspectral imaging (HSI) and video compressive sensing have received significant research attention in recent years. Among existing SCI reconstruction algorithms, deep learning-based methods stand out as their promising performance and efficient inference. However, the deep reconstruction network may suffer from overlarge model size and highly-specialized network design, which inevitably lead to costly training time, high memory usage, and limited flexibility, thus discouraging the deployments of SCI systems in practical scenarios. In this paper, we tackle the above challenges by proposing a simple yet highly efficient reconstruction method, namely stacked residual network (SRN), by revisiting the residual learning strategy with nested structures and spatial-invariant property. The proposed SRN empowers high-fidelity data retrieval with fewer computation operations and negligible model size compared with existing networks, and also serves as a versatile backbone applicable for both hyperspectral and video data. Based on the proposed backbone, we first develop the channel attention enhanced SRN (CAE-SRN) to explore the spectral inter-dependencies for fine-grained spatial estimation in HSI. We then employ SRN as a deep denoiser and incorporate it into a generalized alternating projection (GAP) framework -- resulting in GAP-SRN -- to handle the video compressive sensing task. Experimental results demonstrate the state-of-the-art performance, high computational efficiency of the proposed SRN on two SCI applications.

Compressed Imaging Reconstruction with Sparse Random Projection

Cloud Based Sparse Random Projection for Compressed Imaging.

Image Reconstruction Based on Sub-Gaussian Random Projection

Image reconstruction based on back propagation learning in Compressed Sensing theory

Image Reconstruction Via Small Block-Based Compressed Sensing and Discrete Legendre Polynomials

Compressed Sensing for Image Reconstruction Via Back-off and Rectification of Greedy Algorithm.

Compressed Multi-view Imaging with Joint Reconstruction

The algorithm of image representation and reconstruction based on compressed sensing with composite measurement

Compressed Sensing Image Reconstruction Based On Joint Statistical And Structural Priors

Image Compressed Sensing Reconstruction with 3D Transform Domain Collaborative Filtering

Compressive Sampling and Sparse Reconstruction of Images/videos

Compressed Image Sensing by Jointly Leveraging Multi-Scale Heterogeneous Priors for the Internet of Multimedia Things

Image Compressed Sensing Reconstruction By Collaborative Use Of Statistical And Structural Priors

Compressive Sensing Reconstruction for Compressible Signal Based on Projection Replacement

A Simple and Efficient Reconstruction Backbone for Snapshot Compressive Imaging

Optimization Methods of Compressively Sensed Image Reconstruction Based on Single-Pixel Imaging

Downward Spatially-Scalable Image Reconstruction Based on Compressed Sensing

A Comprehensive Review on Sparse Representation and Compressed Perception in Optical Image Reconstruction

Single Pixel Imaging Via Sparse Projection Angle Sampling

A Multi-Level Residual Reconstruction Based Image Compressed Sensing Recovery Scheme

Progressive versus Random Projections for Compressive Capture of Images, Lightfields and Higher Dimensional Visual Signals