Research On Demosaicking Restoration And Quality Improvement Aigorithm Of Color Images

Most consumer-level digital cameras use a single CCD chip whose surface is covered with a color filter array to obtain color images.The original image has only one color pixel value among the three primary colors of red,green,and blue at each pixel point.In order to get a complete color image,the process of filling and repairing the missing pixels is called demosaicking.It is the core operation of consumer-level digital cameras and an important prerequisite for subsequent digital image processing tasks such as image enhancement,image segmentation,and image recognition.Because of the shortcomings of the traditional interpolation methods that the image details are not recovered well,it is crucial to develop demosaicking algorithm.Based on the typical demosaicking algorithm,this paper proposes a color image demosaicking restoration and quality improvement algorithm for Bayer pattern mosaic images,combined with the current trend of convolutional neural networks.The algorithm is mainly divided into two parts: initial demosaicking and post-detail recovery processing.The main work of this article includes:1.A gradient-based residual interpolation method is proposed.In view of the insufficient use of channel information,the algorithm uses the local correlation between images and the characteristics of guided filtering to reconstruct the green channel,mainly processing the gradient in the horizontal and vertical directions to reconstruct the high-quality green channel.Use the clearer and more complete characteristics of thereconstructed green channel to refine the red and blue channels of the missing pixels,and use residual interpolation and bicubic interpolation to obtain more local information to generate corresponding red and blue channels.Channel pixel value,the reconstructed three-channel pixel value is fused to restore a color image.This algorithm makes better use of the information between channels,and it is found through experiments that the image reconstructed by this method has a certain improvement over the traditional method.2.A lifting algorithm based on convolutional neural network is proposed.In view of the problems of color artifacts and loss of details in the recovered image,the algorithm equates the optimization problem of the wrong pixel value reconstructed from the preliminary reconstructed image to a denoising problem.The difference interpolation method reconstructs the image for optimal processing.What is obtained by training the convolutional neural network is the residual image,that is,the difference between the reconstructed image with errors and the ideal color image.Combining the trained residual image with the reconstructed image with errors to obtain the lifted image.Because the data recorded in the original mosaic image is accurate at each pixel value,and the lifted image and the original data cannot be exactly the same,the optimized image is corrected again with the original data.The algorithm fully combines the interpolation method with the convolutional neural network.The final experimental results show that the algorithm has a good optimization effect and has been significantly improved in image quality and details.