Research On Real-time Bilateral-grid-based Weighted Median Filter And Its Applications

Edge-preserving image smoothing filter is a very important research subject in the field of digital image processing. Its purpose is to remove the noise of the noise image while keeping the image boundary structure information as much as possible, It aims to restore the original image. In the recent years, there are many edge-preserving the image smoothing filters were raised. The standard median filtering method was proved that it can get rid of the salt and pepper noise, but not very good at the edge preserving. The weighted median filtering method is compared with the standard median filtering method can better keep the structure of the boundary information, but need to spend more time. In order to improve the efficiency of weighted median filter method, this paper proprosed a novel weighted median filter algorithm which based on four-dimensional bilateral grid. This algorithm adds a range dimension based on 3D grid of the existinog bilateral, to build a new data structure. namely four-dimensional bilateral grid. The edge-aware weights and the weighted median values are computed in four-dimensional space. This operator greatly speed up the algorithm processing speed. Moreover, the local nature of algorithm enabled an efficient GPU parallel implementation.The GPU-based algorithm can improve the processing speed especially on high-end NVIDIA graphics cards. The effectiveness of the proposed algorithm is illustrated by comparison with state-of-art weighted median filtering methods. Many applications including detail enhancement,image stylization, image transfer style, and image colorization are presented in this paper. These applications demonstrate the effectiveness of the proposed weighted median filter.