<p>The guided image filter is a prevalent image smoothing technique that can be applied to a variety of low-level vision tasks. However, it suffers from the luminance and detail halo artifacts. Existing adaptations of the guided image filter fail to consider these artifacts in a unified framework. In this paper, we propose an edge-aware weighted lasso regression model for guided image filtering. Compared to the ridge regression in the original guided image filter, the weighted lasso regression imposes sparser regularization and incorporates edge-aware weights, thus alleviating the luminance halos. Furthermore, we propose a novel soft erosion scheme to fuse the linear coefficients derived from the weighted lasso regression. The soft erosion scheme reduces the instabilities of the regression around the salient edges, thus being effective in suppressing the detail halo artifacts. Finally, we propose an iterative filtering scheme that further alleviates the luminance and detail halo artifacts. We have conducted experiments across various application scenarios, including image smoothing, HDR tone mapping, JPEG compression artifact removal, texture removal, and detail enhancement, both qualitative and quantitative results suggest the superiority of the proposed filter over the state-of-the-art guided image filters. Furthermore, our filter is efficient, it renders interactive smoothing of 720P 3-channel color images.</p>