MG-CNFNet: A multiple grained channel normalized fusion networks for medical image deblurring

Abstract

In different medical imaging techniques, the relative displacement between the patient and detection instrument can lead to different degrees of motion blur, which will affect the diagnosis of the patient’s condition. In this paper, we propose a deblurring model with Multiple Grained Channel Normalized Fusion Networks (MG-CNFNet), which decomposes the image input by multi-grained division to cascade feature extraction from fine to coarse, constructs the encoder using semi-channel self-normalization to enrich the multi-scale information of the image, highlight the main features and improve the deblurring performance, and uses a shallow convolutional fusion block to obtain more shallow information and balance the performance and efficiency of deblurring. Compared with other state-of-the-art methods, our method is qualitatively and quantitatively evaluated on three medical image datasets (including X-ray, Ultrasound, and MRI) to demonstrate its effectiveness in removing motion blur from medical images, and yields better restoration results.