KEYWORDS: Reconstruction algorithms, Integral imaging, Detection and tracking algorithms, Electron multiplying charge coupled devices, Charge-coupled devices, Image quality, Image processing, 3D acquisition, 3D image processing, 3D displays
Electron-multiplying charge-coupled device (EMCCD) has the characteristic of single photon response under a low-light environment. It is proposed that the reconstruction algorithm of low-light integral imaging by EMCCD reconstruct the details of the target under a low-light environment. First, the algorithm acquires a series of element images by EMCCD integral imaging system. Second, as grayscale values of different element images of the same target meet Poisson distribution, the algorithm introduces a local self-adaptive factor and derives the posterior probability distribution of grayscale value of the target. Finally, it calculates the new element images by posterior probability distribution and reconstructs the target image by updated element images. Experimental results show that the peak signal-to-noise ratio of the reconstructed image by the proposed method is 4.3 dB higher than that of conventional Bayesian estimation. Considering the reconstructed image quality and computational complexity, the overall quality of the reconstructed image is the best when using the 7 × 7 neighborhood range to calculate the local self-adaptive factor in the algorithm. Experimental results show that the proposed algorithm greatly improves the quality of the reconstructed image of the target under low-light environment.
Aiming at the problem of three dimensional reconstruction of photon-limited objects, a new method of Bayesian adaptive estimation is proposed based on the photon counting integral imaging (II) system to improve the quality of reconstructed depth slice images. Firstly, an array of photon counting elemental images is obtained by the photon counting II system. Then, a local adaptive mean factor is introduced into the Bayesian framework as a form of exponential prior distribution. Finally, elemental images estimated by the posterior mean are back propagated to reconstruct depth slice images. The experimental results reconstructed by the proposed method achieve higher peak signal-to-noise ratio than the traditional Bayesian method under photon-starved conditions.