Blind Measurement of Image Blur for Vision-Based Applications

Shiqian Wu,Shoulie Xie,Weisi Lin
DOI: https://doi.org/10.1007/978-3-642-19551-8_6
2011-01-01
Abstract:This chapter presents a novel metric for image quality assessment from a single image. The key idea is to estimate the point spread function (PSF) from the line spread function (LSF), whereas the LSF is constructed from edge information. It is proven that an edge point corresponds to the local maximal gradient in a blurred image, and therefore edges can be extracted from blurred images by conventional edge detectors. To achieve high accuracy, local Radon transform is implemented and a number of LSFs are extracted from each edge. The experimental results on a variety of synthetic and real blurred images validate the proposed method. To improve the system efficiency, a criterion for edge sharpness is further proposed and only the edge points from sharp edges are selected for extracting the LSF without using all edge information. The effects of nearby edges on the selected edge feature and the resultant LSF are analyzed, and two constrains are proposed to determine appropriate LSFs. The experimental results demonstrate the accuracy and efficiency of the proposed paradigm. This scheme has fast speed and can be served in blind image quality evaluation for real-time automatic machine-vision-based applications.
What problem does this paper attempt to address?