An Improved Guided Filtering Algorithm for Polarized Images by Using LOG Operator

Guided image filter (GIF) is a commonly adopted local filter in image processing and computer vision, due to its edge-preserving property and low complexity. However, GIF may suffer from halo artifacts. In this paper, we present an improved guided filtering algorithm by using Laplacian-of-Gaussian (LOG) operator, which is dedicated to division-of-focal-plane (DoFP) polarization images to reduce the edge halo artifacts. By calculating the magnitude response of the LOG operator in the local window, the initial regularization parameter becomes self-adaptive to the values of the smooth area and edge area of the image. It can highlight the difference between edge pixels and the pixels in flat areas to show more details, especially at the edge. Experimental results demonstrate that the proposed algorithm performs better than the reference method in terms of both visual effect and root mean square error (RMSE). Especially for degree of linear polarization (DoLP) images, the RMSE has been significantly reduced by around 88%. In addition, the proposed filter made these improvements without computational overhead.