基于深度学习的单像素相机超分辨率成像优化

Super-Resolution Imaging Optimization of Single Pixel Camera Based on Deep Learning

摘要:

基于压缩感知和单像素成像的基本原理,设计了一种用于图像超分辨率重建的新型深度卷积神经网络架构。这种单像素超分辨率成像算法成功地将深度学习图像超分辨率重建技术与压缩感知单像素成像技术相结合,从而发展出一种全新的深度学习单像素成像优化方法。与传统的常规压缩感知图像重构算法相比,该算法有效提升了图像超分辨率重建精度和单像素成像质量。通过图像重建的仿真实验和单像素相机的成像实验验证,结果表明这种基于深度学习的新型单像素相机成像方式具有良好的性能表现。

Abstract:

Based on the basic principles of compressed sensing (CS) and single pixel camera imaging, a new deep convolutional neural network architecture for image super-resolution reconstruction is redesigned. The new single pixel super-resolution imaging algorithm successfully combines deep learning image super-resolution reconstruction technology with compressed sensing single pixel imaging technology to develop an entirely new deep learning based optimized imaging method of single pixel camera. Compared with the conventional CS-based algorithms, the accuracy of image super-resolution reconstruction and the imaging quality of single pixel camera are improved effectively. The good performance of the new proposed deep learning imaging method of single pixel camera has been verified by simulation experiments of image reconstruction and imaging experiments of single pixel camera.