No-reference blur assessment based on edge modeling

• We propose a no-reference blur metric by integrating edge modeling into JNB.
• A parametric edge model is introduced for edge description and computation.
• We advocate using the salient edge pixels to simulate the blur assessment of HVS.
• With the pixel-adaptive width and contrast, the metric is simple yet accurate.

This paper presents a no-reference objective blur metric based on edge model (EMBM) to address the image blur assessment problem. A parametric edge model is incorporated to describe and detect edges, which can offer simultaneous width and contrast estimation for each edge pixel. With the pixel-adaptive width and contrast estimations, the probability of detecting blur at edge pixels can be determined. Also, unlike previous work, we advocate using only the salient edge pixels to simulate the blur assessment in Human Visual System (HVS). Finally, the blur metric is obtained by cumulating the probability of blur detection. Various images with different blur distortions are tested to demonstrate the effectiveness of the proposed metric.