A noise-immune no-reference metric for estimating blurriness value of an image

• We propose a simple yet accurate no-reference blur metric with low computational cost which is robust against noise.
• The proposed blur metric is based on the observation that there is a considerable difference between the DCT of a sharp image and the one associated with its blurred version.
• The experiments, performed on four databases (including CSIQ, TID, IVC, and LIVE), indicate the capability of the proposed metric for measuring the amount of blurriness in images, especially at the presence of noise.

Blur is a type of distortion that may happen in digital images. Blur estimation is an important issue in image processing applications such as image deblurring and depth estimation. Several blur metrics exist in the literature, but they are mostly sensitive to the presence of noise. In this paper, a simple yet accurate no-reference blur metric with low computational cost is proposed, which is robust against noise. The proposed blur metric is based on the observation that there is a considerable difference between the DCT of a sharp image and the one associated with its blurred version. The effect of noise is mainly reflected in the higher order DCT coefficients. Hence, the noise effect is mitigated in this paper via discarding the higher order DCT coefficients. The experiments, performed on four databases (including CSIQ, TID2008, IVC, and LIVE), indicate the capability of the proposed metric in measuring image blurriness. Comparative results with other existing approaches show the superiority of the proposed blur metric, especially at the presence of noise.