검색 상세

A Piecewise-Linear PSF Model for Motion Deblurring

움직임에 의한 영상 번짐 제거를 위한 부분선형 PSF 모델

초록/요약 도움말

Under low lighting conditions, noise and blur can be major components of image degradation. While they are usually separately dealt with in image enhancement algorithms, they are tightly related in the sense that higher ISO values and longer exposure time ensure brightness of images at the cost of noise and blur, respectively. Therefore, images acquired in dark scenes usually contain noise and blur at the same time. Estimation of point spread function (PSF), which characterizes blurring in image, is a crucial step in the deblurring process. The estimation is easily perturbed by noise and estimation errors introduce more pronounced artifacts in deblurred images. Thus, in this dissertation, we propose a new model for the PSF based on a parametric curve representation. Since every motion blur can be represented by an arbitrary parametric curve, a piecewise-linear curve to approximate the curve using a small number of free parameters. In the perspective of the PSF estimation, DOF (degrees of freedom) of the PSF model plays an important role in expressiveness and noise robustness. The model is found to be an effective tradeoff between expressiveness and robustness as it takes advantage of two extremes: the generic model represented by a discrete 2-dimensional function with high DOF for maximum flexibility but suffering from noise, and the linear model which improves robustness and simplicity but has limited expressiveness due to its low DOF. Several deblurring methods based on the generic model and the piecewise-linear model as an alternative were tested. After analyzing the experiment results based on real-world images with significant levels of noise and a benchmark data set, we conclude that the proposed model is not only robust with respect to noise, but also flexible in addressing various types of blur.

more