Single-band spectral light field images reconstruction based on compressed sensing

摘要

For solving the difficulty of acquiring single-band spectral light field images, the spectral light field images using compressed sensing based on overcomplete dictionary is proposed and simulated. Firstly, multiple sets of light field images are collected as a sample set for training overcomplete dictionary, and the overcomplete dictionary is generated by training the sample set. Then, using random matrix of periodic arrangement as the measurement matrix to realize the reduced-dimensional sampling of the signal. Finally, single-band spectral light field images are reconstructed using compressed sensing reconstruction algorithm. The peak signal to noise ratio of the reconstructed images is 30.5 dB. The experimental results show that the spectral light field image reconstructed by this method has sufficient parallax and can recover the spectral information carried by the image. And the sampling rate of this method proposed is only 4% of the image size. This method proposed effectively solves the problems that the image capturing process of the spectral light field is complicated and the data volume is large. This method provides a new way to reconstruct high-resolution single-band spectral light field images with low sampling rate, and provides experimental data and new ideas to further increase the depth information of spectral images for generating 3D spectral images.