Non-quadratic Regularization Based Image Deblurring: Automatic Parameter Selection and Feature Based Evaluation

In computer vision based analysis, a completely automatic inspection of parts on assembly line involves many challenges. Since the parts are moving fast on line it is most probable that the captured frames are motion blurred and noisy images. Therefore accurate extraction of features from the image may not be possible. To overcome this challenge, we consider quadratic and non-quadratic regularization based deblurring. To select the regularization parameter automatically, we propose usage of unbiased predictive risk estimator method. We investigate the quantitative effect of the applied methods on feature extraction performance and demonstrate the effectiveness of the proposed approach with experiments on real data.