A No-Reference Image Blur Metric

This paper presents a no-reference image quality measure, targeted towards blur distortions based on the study of human blur perception for varying contrast values. Utilizing sensitivity of human blur perception at different contrasts, a probabilistic framework is developed. This framework is used to estimate the probability of detecting blur at each edge in an image. The blur perception information at each edge is then pooled over the entire image to obtain a final quality score by evaluating the cumulative probability of blur detection. Proposed metric is able to predict relative amount of blurriness in images. Higher metric value represent less blurred image. Results are provided to illustrate the performance of proposed metric. Performance of proposed metric is compared with existing no reference image quality metric for various publically available image databases.