Depth free entropy based ghost removal on synthetic aperture images

The depth resolution of synthetic aperture photography (SAP) is significantly improved than traditional imaging system. However, as one side effect, ghost artifact is brought in defocused regions of SAP images. In this paper, we propose an entropy based ghost removal algorithm on SAP images, which is free of depth estimation. First, we analyze the essence of ghost artifact from light field. Then we present three entropy based assumptions for predicting the focus trend and estimating the kernel of low pass filter for each pixel. Based on the neighborhood pixel similarity and adjacent frame continuity, a low pass filter is employed to reduce the unexpected ghosting in the spatial domain. Experimental results have verified the validity of proposed algorithm on both virtual and real data. The ghosting areas are greatly smoothed while focused regions are kept as same as original image.