The quality of real-world image or video is usually corrupted by blur and noise.Detecting the degree of the degradation is the vital preconditions for many computer vision issues.In this paper,we focus on the local perceptual-based blur and noise detection adaptively.For the image noise detection,we propose a perceptual-based full-reference metric and a no-reference metric by integrating perceptually weighted local noise into a probability summation model.Experiments are conducted on two classical databases: LIVE and TID2008.Compared with other quality metrics,our metric achieve much better performance across different type of noise.The proposed method can also predict image quality across different content and different noise more precisely.In order to deal with issues of spatially-varying blur and time-consuming problem,we take human perception into our account.By integrating probability of blur detection,the proposed metric generate a blur map indicating the level of local blur,which also use the directional edge spread calculation and local probability summation.For the problem of flat regions detection,Just Noticeable Difference is further integrated to generate perceptually significant blur maps.Compared with other blur detection methods,experimental results indicate that the proposed method manifests better performance quantitatively. |