Abstract
In the past decade, there has been an increasing success of guided image filtering (GIF). Leveraging the guidance image as a prior and transferring the structural details to the target image, GIF has demonstrated its ability in faithfully preserving image edges while maintaining low computational complexity. Additionally, GIF exhibits good capability in extracting and characterizing images from various domains. Researchers have proposed large numbers of GIFlike variants. Nevertheless, limited effort has been devoted to a systematic review and evaluation of these methods. To fill this gap, this paper provides a comprehensive survey of existing GIF-like methods, including model-and deep learning-based approaches. Moreover, extensive experiments are conducted to compare the performance of 18 representative methods. Analysis of the qualitative and quantitative results reveals several observations concerning the current state of this area. To address the challenge of content-blindness, Guided Image Filtering (GIF) [7] has received extensive attention from the research community. The key idea of GIF is to utilize an additional image as guidance to transfer the structure information of the guidance to the degraded target image, thus allowing the restoration of blurred edges or suppression of noise. This provides a new way to view the filtering process and has a wide range of applications. As shown in Table 1 , we can categorize the applications into two classes based on whether the guidance and target images are derived from the same modality: self-guidance based and reference-guidance based tasks. However, two key assumptions of GIF, namely the locally linear model and the structure consistency, are often violated in certain scenes, leading to halo and texture-copy artifacts, respectively. Specifically, locally linear model assumes a local linear relationship between the output and guidance image. As a result, halo artifacts may occur around enhanced edges when the edges are blurred in the smoothed image, as illustrated in the top row of Fig. 1 . Moreover, structure consistency between target and guidance images is often challenged when the two input streams are in different modalities (e.g., RGB and depth [10] ), which can result in texture-copy visual artifacts, as shown in the bottom row of Fig. 1 . Motivated by the remarkable success of GIF and its two limitations, researchers have devoted much attention to this field. A Table 1 : Examples of applications based on whether guidance and target images stem from the same modality.
Showing the abstract — retrieve the full paper via the Exa API.