Research On Image Denoising And Fusion Based On Joint Bilateral Filtering

The complexity of the human visual system and the rapid development of the digital image processing technology that put forward new requirements and challenges in traditional image processing and computer vision tasks,especially the low level image fusion and image noise tasks.Due to interference of external noise or multi-sensor image acquisition,the image information is not comprehensive or reliable,which reduces the image quality and restricts the performance of subsequent image processing tasks.The core problem of image denoising is how to efficiently remove the noise and preserve image edge,contour and texture detail information.As an important method of image denoising and fusion,bilateral filter has low time complexity,and can keep the edge information well when perform the image denoising.Recently,a large number of joint bilateral filtering algorithms and image denoising and multi-scale decomposition algorithm based on bilateral filtering have emerged.However,many existing joint bilateral filtering will lead to image blurring in image noise suppression,and the limitation of the noise type,in addition to bilateral filtering in image decomposition process,for failing to adequately acquire edge information which leads to the loss of the fusion image structure and the heavy artifacts.In this thesis,we proposed two novel low level image processing methods based on the joint bilateral filtering: based on SURE(Stein 's unbiased risk estimate)method to estimate the edge weighted SEWBF(SURE estimator for bilateral filtering Edge-preserving Weighted Bilateral Filtering),and FCGBF image fusion algorithm based on gradient cross bilateral filter(Image Fusion based on Cross-gradient Bilateral Filtering)to obtain high resolution to preserve the integrity of the structure,image denoising and image fusion.The details of the above methods are described as follows:(1)The SEWBF algorithm is a weighted filtering algorithm,which possesses the comprehensive advantages of pseudo median bilateral filtering and robust improved bilateral filtering,can achieve robustness and maintain the edge effect.Moreover,the inhibitory effect of Gauss noise and salt pepper noise and other noise types are very good.In this thesis,a novel robust kernel function is proposed,which is then applied to the pseudo median bilateral filter which can effectively remove salt and pepper noise.At the same time,the robust bilateral filter is introduced,which involves the estimation of the mean square error,and then the SURE estimation method is used to define the robust bilateral filtering denoising model to obtain the optimal parameters.Finally,the optimal weight factor is calculated by SURE,and the SEWBF algorithm is obtained by weighted mean of improved pseudo median bilateral filter and robust bilateral filtering.Finally,the effectiveness of the SEWBF algorithm is verified on multiple sets of experiments,and the experimental results are analyzed.(2)The FCGBF algorithm is a kind of fusion method for different types of source images,which is an improvement on the image fusion algorithm based on the cross bilateral filter.The FCGBF algorithm uses the improved gradient cross bilateral filtering to decompose the image,in order to improve the accuracy of image decomposition.By using gradient function instead of gray similarity kernel function,it can get the edge gradient information and reduce interference artifacts,and finally improve the algorithm's performance.Then,in order to make the fusion results retain more of the source image information while reducing artifacts,we employ an adaptive fusion rule of decomposition of information fusion to calculate the horizontal and vertical edge strength in the calculation of the weight in the process of fusion.Finally,experiments are carried out to demonstrate the effectiveness of FCGBF,such as multimodal,multi focus,medical images and so on.