Anisotropic Gaussian Side Windows Guided Filtering
Fuping Wang, Congcong Ji · 2021 4th International Conference on Artificial Intelligence and Pattern Recognition · 2021
The guided filtering and its different variants have good edge preserving characteristics, widely used in many computer vision applications such as Matting, Dhaze, but halos still appear near the edges and it is difficult to identify high-resolution edges. Therefore, we propose an Anisotropic Gaussian Side Window Guided Filtering (ANGKSWGF). This method has three advantages: (1) We efficiently combine Anisotropic Gaussian-shaped Kernel with side window technology to suppress edge blur and halo effects; (2) We use anisotropic Gaussian-shaped kernel weights The factor effectively weights and fuses different pixel information in the window to produce more accurate guided filtering results; (3) The multi-directional anisotropic Gaussian kernel with large-scale shape parameters can effectively improve the enhancement effect of fine edges. The proposed filtering algorithm can be applied to image processing applications, such as image smoothing and detail enhancement. The experimental results on synthetic ideal edge and actual natural images show that the edge preservation and detail enhancement of the proposed algorithm is significantly better than the other algorithms.