Histogram modelling-based no reference blur quality measure

•A normalized, simple, fast, and faithful objective blind blur quality measure is proposed.•The Human Visual System (HVS) is sensitive to blur effect especially on high frequency components corresponding to edges and texture.•The histogram of the absolute quantified DCT coefficients of the edge map can be modelled by using an exponential Probability Density Function (PDF).•The statistical parameter of the exponential PDF is used as a cue to characterize the blur effect.

Blurring is a common artefact detrimental to the image quality. It affects especially edges and texture features that represent high frequency components of an image. The purpose of this paper is to propose a simple, fast, and faithful measure able to blindly assess blur amount in images. The main idea turns on analysing the frequency response at the multiresolutiontransitions. To achieve that, the histogram of the discrete cosine transform coefficients of the edge map is modelled by using an exponential probability density function (pdf). Tests revealed that the steepness of the pdf depends on the blur amount, hence, it is used as a cue to characterize the blur effect. Comprehensive testing demonstrates good consistency of the proposed measure with subjective quality scores as well as satisfactory performance when compared with representative state-of-the-art blind blur quality measures.