ReconNet: Non-Iterative Reconstruction of Images from Compressively Sensed Measurements

Abstract

Ground Truth 25% measurements 4% measurements 1% measurementsFigure 1: Given the block-wise compressively sensed (CS) measurements, our non-iterative algorithm is capable of high quality reconstructions. Notice how fine structures like tiger stripes or letter 'A' are recovered from only 4% measurements. Despite the expected degradation at measurement rate of 1%, the reconstructions retain rich semantic content in the image. For example, one can easily see that there are two tigers resting on rocks, although the stripes are blurry. This clearly points us to the possibility of CS based imaging becoming a resource-efficient solution in applications, where the final goal is high-level image understanding rather than exact reconstruction. AbstractThe goal of this paper is to present a non-iterative and more importantly an extremely fast algorithm to reconstruct images from compressively sensed (CS) random measurements. To this end, we propose a novel convolutional neural network (CNN) architecture which takes in CS measurements of an image as input and outputs an intermediate reconstruction. We call this network, ReconNet. The intermediate reconstruction is fed into an off-the-shelf denoiser to obtain the final reconstructed image. On a standard dataset of images we show significant improvements in reconstruction results (both in terms of PSNR and time complexity) over state-of-the-art iterative CS reconstruction algorithms at various measurement rates. Further, through qualitative experiments on real data collected using our block single pixel camera (SPC), we show that our network is highly robust to sensor noise and can recover visually better quality images than competitive algorithms at extremely low sens-ing rates of 0.1 and 0.04. To demonstrate that our algorithm can recover semantically informative images even at a low measurement rate of 0.01, we present a very robust proof of concept real-time visual tracking application.

Machine Learning Based Single-Frame Super-Resolution Processing for Lensless Blood Cell Counting

“…Recently, another category of machine learning based SR approaches is developing quickly [17,18,19,20,21,22,23,24,25]. Machine learning has very good performance and applications on a variety of problems such as visual/speech recognition, natural language processing, and biomedical imaging, etc.…”

Parallel imaging and convolutional neural network combined fast MR image reconstruction: Applications in low‐latency accelerated real‐time imaging

“…As a result, image reconstructions using total variation as regularizations could result in oversmoothed images with blurred boundaries between structures. Inspired by early work of using sparse dictionary learning‐based regularization for MR reconstruction, and more recent work of using convolutional neural networks (CNN) for natural image reconstruction, we propose to use a general CNN‐based regularization in this work. Specifically, assume maps an artifact‐contaminated image, mainly caused by k‐space undersampling and results in structural ghosting aliasing or incoherent noise‐like artifacts on image, to an artifact‐free image.…”

“…Specifically, assume maps an artifact‐contaminated image, mainly caused by k‐space undersampling and results in structural ghosting aliasing or incoherent noise‐like artifacts on image, to an artifact‐free image. It can be represented by multiple layers of convolutional kernels parameterized by . Hence, a compact representation of this mapping is:…”

“…(), which is a pure quadratic optimization problem, can be obtained by setting its derivative to be zero, and solving a system of linear equations through direct matrix inversion or pseudo‐inversion, we chose to solve it with the gradient descent algorithm due to the large size of the coefficient matrix () when multiple coil sensitivity maps are included, as well as its poor conditioning when undersampling factor is high. For previously mentioned natural image reconstruction task, learning parameters for is usually performed in image domain only. However, by incorporating the k‐space data in training, which is a representation of the image in the spatial‐frequency domain, we hypothesize that we will be able to improve image reconstructions compared to traditional image‐based training in terms of image artifact removal when the image was acquired with k‐space undersampling.…”

Single‐shot T2 mapping using overlapping‐echo detachment planar imaging and a deep convolutional neural network

“…Deep neural network (DNN) architectures, convolutional neural network (CNN) in particular, are finding more and more applications in medical imaging analysis for various problems including classification, detection and segmentation . It has already been demonstrated that CNN outperforms sparsity‐based methods in super‐resolution reconstruction …”