虽然基于像素重排列的迭代反投影算法已经在TOMBO模型构建时提出,但是该方法需要大量的迭代次数,同时在噪声平滑效果上还有待于改进.因此一种正则化的迭代反投影算法被提出为该系统重构图像.采用自适应的总变差正则化因子和双边总变差正则化因子来正则化迭代反投影算法.自适应总变差正则化因子根据图像的当前信息来选择参数,因此用该因子正则化后的迭代反投影算法可以在平滑噪声的同时保留高频成分.而双边总变差正则化因子是依据像素点的最邻近领域和次邻近领域来判别该点是否为噪声点,考虑了更多的图像信息,从而可以跨过边缘平滑噪声.同时双边总变差正则化因子可以大大地加速重构的过程.实验是建立在仿生复眼图像上,实验结果证明了这两种正则化的迭代算法的有效性.%Iterative back-projection algorithms based on pixel-rearrangement have been reported since thin observation modules using bound optical(TOMBO) systems were presented. However computational costs are high and the noise smoothing ability needs to be improved. So a normalized iterative back-projection algorithm was proposed to get reconstructed images for this system. Both an adaptive total variation normalized factor and a bilateral total variation normalized factor were employed to normalize the iterative back-projection algorithm. The former factor chose parameters adaptively according to present image information, so it could retrieve the high frequency components while smoothing noise. However, the latter factor determined the noise points according to nearest and next nearest neighborhood of pixels, in which more image information was considered so that it smoothed noises across boundaries. Also the latter factor accelerated the reconstructing process greatly. Experiments were based on bionic compound eye images. And the results demonstrated that the suggested method requires less time in image reconstruction.
展开▼
