The Edge-preserving Improvement And Applications Of Guided Image Filtering

Currently,digital image plays a vital role in many fields,such as transportation,mobile payment,agricultural protection,biomedicine,urban security,aerospace,weather forecast and industrial detection,etc.However,due to the limitations of imaging equipment,environment and transmission condition,image is often contaminated by noise,weather or light in the acquisition process,resulting in quality degradation,which makes it hard to be applied to the subsequent tasks directly.Guided image filter,as a local linear edge-preserving smoothing operator to improve image quality,has attracted great attentions of researchers since its birth,and has been widely used in image denoising,detail enhancement,dehazing,super-resolution and image fusion,etc.However,this filter still suffers from the defects of halo artifacts,over smoothing and texture copy.This paper studies these drawbacks profoundly.The main contents and contributions are the follows:1)Exploring the edge-preserving mechanism of the prevalent image processing methods,especially for bilateral filter,joint bilateral filter,steering kernel regression,guided image filter and its variants;2)Proposing a novel guided image filtering method based on steering kernel,and demonstrate the effectiveness of it to suppress halo artifacts and improve edge-preserving performance with respect to theoretical analysis and experimental results;3)Proposing the guided image filtering with collaborative guidances,which takes both structure similarity and edge-strength between the target and guidance images into consideration,and address texture copy and edge blur in existing guided filters simultaneously.