An Automatic Blind Image Deblurring Method via Label Confidence
Ming Fan, Ming Ma, Guanghao Jin · 2020 IEEE 4th Information Technology, Networking, Electronic and Automation Control Conference (ITNEC) · 2020
Image deblurring has been an active research topic in the computer vision and image processing. Generally, the existing deep learning methods can get high quality deblurred image while the quality depends on the variety of training dataset for blurred images. On the other side, the blur kernel estimation methods can work well on wider range images without training dataset for blurred images while the quality depends on the manual tuning of parameters. Label confidence is the production of classification result by deep learning method. We found that the image is in the process from blur to clear, the label confidence of the image increases. In our paper we introduced a new blur kernel estimation method that uses label confidence and Laplace variance to perform auto tuning, so that it can select the best deblurred image from the outputs of the blur kernel estimation method automatically. The experimental results show that our method can achieve automatic image deblurring and generate higher quality images than the existing state-of-art deblurring method.