A no-reference infrared image sharpness assessment based on singular value decomposition

Recent image sharpness metrics are usually proposed for visible light image and infrared image sharpness assessment is seldom discussed. As for infrared images, we usually concern more about the salient regions. Therefore, in this paper, a novel no-reference algorithm based on saliency detection (SD) and singular value decomposition (SVD) is proposed to assess infrared image sharpness. Gaussian blur is first used to build a reference image. Then salient regions are detected by combining the local mean and variation. Next, singular value decomposition-based metric is proposed to evaluate the variation between original image and reference image. The image quality score is finally obtained by using the five-parameter logistic regression. Experimental results show that the proposed method correlates well with the subjective quality evaluations of infrared images and is highly competitive with state-of-the-art visible light image sharpness metrics.