The Research On Image Deblurring Algorithm Based On Reference

Image blur is usually due to entangled motion of objects in the captured scene or camera shake.Image deblurring has been a challenging problem in computer vision and image processing,where the goal is to recover an image with sharp details from a given blurred image.Blurry images not only affect people’s visual perception quality,but also reduce the performance of visual tasks such as object detection,face recognition,image segmentation,etc.Therefore,although image deblurring is a low-level computer vision task,it is of great significance to study an effective deblurring algorithm to restore image structure and texture.However,most of the existing image deblurring methods use a single blurred image as the input of the algorithm,which limits the information acquisition and fails to preserve satisfactory structures and textures.Aiming at the deficiencies of existing image deblurring algorithms,this paper uses deep learning technology to conduct in-depth research.The work of the article mainly includes the following parts:(1)A new method for image deblurring with the help of a clear reference image is proposed,which uses the reference image to obtain high-quality deblurring results.To make better use of clear reference images,we develop an encoder-decoder network and design two novel modules to guide the network to better recover images.The proposed Reference Extraction Aggregation Module can effectively establish the correspondence between blurred images and reference images,and explore the most relevant features for better blur removal,and the proposed Spatial Feature Fusion Module enables the encoder to perceive blur information at different spatial scales.Finally,the multi-scale features from the encoder and cascaded Reference Extraction Aggregation Module are integrated into the decoder for global fusion and representation.Extensive experimental results on different benchmark datasets demonstrate that the method outperforms both quantitatively and qualitatively.(2)A reference-based dual-task framework is designed for image deblurring by deblurring and refining blurred images under the guidance of reference images to obtain highquality images.Specifically,the framework includes two tasks of single image deblurring and reference feature transfer.The single image deblurring task deblurs the blurred image using only the blurred image itself.The reference feature transfer task extracts rich texture features from a reference image and transfers them to the coarse results of a single image deblurring task.Benefiting from the reference images,the proposed method achieves more realistic visuals and sharper texture details.Experimental results on GoPro,HIDE and RealBlur datasets show that our method outperforms the state-of-the-art methods both quantitatively and qualitatively.