Xiaodong Gu

No Reference Block Based Blur Detection

Blur is one of the most important features related to image quality. Accurately estimating the blur level of an image is of great help to estimate its quality. In this paper, a No Reference Block-based Blur Detection (NR-BBD) algorithm is proposed. It calculates the local blur at the boundaries of Macro Blocks (MBs) and then averages all of them to get the blur of the image. A content dependent weighting scheme is employed to reduce the influence from the texture. Compared with traditional edge based blur metrics, NR-BBD has a lower complexity, exhibits more stable for different image content, and results in a higher correlation with the perceived subjective visual quality (the resulting Pearson Correlation is 0.85 in the data set with 1176 images with different content type and different quality level.).