A modified joint trilateral filter based depth map refinement method

Abstract

We present a modified joint trilateral filter based method for depth map refinement. Seeded region growing algorithm and Euclidean distance transform(EDT) are involved for error pixels extraction. With the error information wiped out, a joint trilateral filter method is employed as a hole filling strategy to fill the values. Compared to the basic bilateral filter method, structural similarity index(SSIM) is added as the third lateral, while a threshold and an aberration condition are involved for modification to protect the edges and improve the confidence of the filling values. Both qualitative analysis and quantitative analysis are shown to testify the performance. Comparisons between our method and the other popular methods are provided.