A Two-Branch Convolution Residual Network for Image Compressive Sensing

Abstract

Deep learning has made great progress in image compressive sensing (CS) tasks recently, and several CS models based on it have achieved superior performance. In practice, sensing the entire image requires huge memory and computational effort. Although the block-based CS method can effectively realize image sensing, it will cause block effects that severely decrease the reconstruction performance. To this end, this paper proposes a two-branch convolution residual network for image compressive sensing (denoted as TCR-CS), which mainly consists of a two-branch convolution autoencoder network and a residual network. Specifically, the two-branch convolution autoencoder network senses the entire image through multiple scale convolutional filters to obtain measurements. For better CS reconstruction, the image is preliminarily reconstructed by the deconvolution decoder network, and then the residual network is used to optimize the pre-reconstructed image. Through the end-to-end training, all networks can be jointly optimized. Finally, experimental results demonstrate that the proposed TCR-CS method is superior to existing state-of-theart CS methods in terms of structural similarity, reconstruction performance and visual quality at different measurement rates.

A Deep Unsupervised Residual - Convolutional Auto-Encoder: Multifocus Image Fusion based on Local Binary Pattern

“…Chenquan et al [28] proposed a two branch CNN that introduces a fusion of images based on compressive sensing and a residual decoder network is used for the reconstruction of the fused image. The basic idea of Boyuan et al [24] , Jin et al [27], and Chenquan et al [28] motivates our proposed work that the residual part is included in both the encoder and decoder part such that it carries over all the feature maps throughout the network.…”

“…Chenquan et al [28] proposed a two branch CNN that introduces a fusion of images based on compressive sensing and a residual decoder network is used for the reconstruction of the fused image. The basic idea of Boyuan et al [24] , Jin et al [27], and Chenquan et al [28] motivates our proposed work that the residual part is included in both the encoder and decoder part such that it carries over all the feature maps throughout the network. From the inspiration of SESF-Fuse: Boyuan et al [24], we also discarded the fusion during the training process but we have utilized both encoder and decoder features for fusion and reconstruction of the fused image.…”

Total Variant Based Average Sparsity Model With Reweighted Analysis for Compressive Sensing of Computed Tomography

“…The framework is a three-stage end-to-end training system where the primary two stages are employed for the reconstruction followed by the performance enhancement stage [27]. Furthermore, a two-branch convolution residual network that is comprised of a two-branch convolution auto-encoder network and a residual network is proposed for CS [28]. Moreover, generative adversarial neural networks are explored in detail manner for the reconstruction of images as CS approaches [29], [30].…”

An Iterative Method With Enhanced Laplacian- Scaled Thresholding for Noise-Robust Compressive Sensing Magnetic Resonance Image Reconstruction

“…Much research on MRI has focused on the reduction of the acquisition time and accurate reconstruction from highly undersampled k-space data, which is a pair of contradictions. Compressive sensing (CS) methods [2,3] as a fundamental developed methodology in information society are able to achieve accurate image reconstruction from very few linear measurements, and have been successfully applied to MRI, which is known as CS-MRI [4,5]. Using CS-MRI techniques one can significantly reduce the amount of kspace data and corresponding acquisition time by means of undersampling without having to sacrifice the quality of reconstructed MR images.…”