Evaluation Of No-Reference Models To Assess Image Sharpness

In the past decades, massive attention has been paid toward no-reference or blind image sharpness assessment (BISA) and many algorithms have achieved good performance. This paper provides an evaluation of 12 state-of-the-art BISA methods based on Gaussian blurring images collected from four simulation databases (LIVE, CSIQ, TID2008 and TID2013). The prediction performance is estimated with two metrics after four or five-parameter non-linear score fitting. Experimental results indicate that the algorithm RISE achieves the best performance. Additionally, the effect of different non-linear scoring fitting methods on the performance evaluation is insignificant. In general, RISE is a visible and significant milestone for BISA algorithm development at present and the future work might be toward novel and real-life applications.