Interactive poisson photometric propagation for facial composite

In image composition, the inconsistent illumination of the source images is one of the major problems for seamless stitching of separated patches. The Poisson image editing is a sound technique for seamless image composition. In this paper, we have generalized and improved this technique and applied it onto solving the illumination discontinuity problem for facial image composition. Toward stitched image with patches of arbitrary shapes, number, and severe photometric discrepancy, an extended Poisson equation is proposed and formulated into a linear equation problem. To solve this equation efficiently, a layer-based Poisson solution propagation algorithm is designed. Based on it, an interactive photometric alignment system for facial compositing image is built. In the experiments, the photometric propagation effects with respect to the standard Poisson editing and other relevant algorithms are compared. Its time performance is also investigated. The experimental results verified the effectiveness and efficiency of the proposed method.