Abstract:As an important naked eyes three-dimensional display technology, integral imaging technology completely records the three-dimensional scene information while the huge amount of data puts pressure on the transmission and storage. In order to achieve effective compression and reconstruction of images, according to the characteristics of photon counting integral imaging, a scheme for image compression and reconstruction based on the distributed compressed sensing theory was proposed. In this scheme, images were divided into reference images and non-reference images, which were set with different measurement rates and reconstructed respectively. To ensure the reconstruction quality of non-reference images, a joint reconstruction algorithm was proposed. Firstly, the block measurements were performed on the non-reference images, and the image blocks classification was carried out by considering the correlation with the reference images. Then, combined the measurement information of the reference images, a new measurement vector was constructed to complete the initial image reconstruction. In order to further improve the image reconstruction quality, the image was reconstructed with secondary residual compensation to obtain the final reconstruction results. Finally, a large number of experiments were conducted by setting different measurement rates. The experimental results show that the image reconstruction quality of the proposed algorithm can reach 30 dB when the measurement rate is0.25, and can reach 35 dB when the measurement rate is 0.4. The performance of the algorithm has certain superiority.

Single Photon Counting Compressive Imaging Based on a Sampling and Reconstruction Integrated Deep Network

Deep-learning-based Single-Photon-counting Compressive Imaging Via Jointly Trained Subpixel Convolution Sampling.

Photon Counting Correction Method to Improve the Quality of Reconstructed Images in Single Photon Compressive Imaging Systems.

Single-Photon Compressive Imaging Based on Residual Codec Network

Convolutional Neural Network Based Computational Imaging Spectroscopy

Single Photon Counting Compressive Imaging Using a Generative Model Optimized Via Sampling and Transfer Learning.

Image reconstruction based on back propagation learning in Compressed Sensing theory

Compressed Sensing Fourier Single Pixel Imaging Algorithm Based on Joint Discrete Gradient and Non-Local Self-Similarity Priors

Fast High-Resolution Imaging Combining Deep Learning and Single-Pixel Imaging

Adaptive Sampling for Image Compressed Sensing Based on Deep Learning

Single Photon Compressive Imaging with Enhanced Quality Using an Untrained Neural Network.

Content-Adaptive Image Compressed Sensing Using Deep Learning

3D compressive imaging system with a single photon-counting detector

Single-shot Compressed Ultrafast Photography Based on U-net Network

Single-photon compressive imaging with some performance benefits over raster scanning

Cascaded Reconstruction Network for Compressive image sensing

High-efficiency Single-Photon Compressed Sensing Imaging Based on the Best Choice Scheme.

Fully Convolutional Measurement Network for Compressive Sensing Image Reconstruction

Compression and Reconstruction of Photon Counting Integral Imaging

Design and Training of Anti-Noise Reconstruction Network for Single-Photon Compression Imaging

Adaptive High-Resolution Imaging Method Based on Compressive Sensing