Research On Image Denoising Algorithm Based On Improved Bemd And Its Application

The advancement of modern computer technology has greatly promoted the research and development of image processing technology,attracting a large number of research scholars to participate in the research field of image processing.The image quality can seriously affect the results of image processing.Therefore,image denoising is a basic and important research topic.Rich edge texture information is contained in images,and the gray-scale mutation along the vertical direction of the texture makes the image have non-stationary signal characteristics,which can be processed by non-stationary signal analysis.The empirical mode decomposition method is an adaptive processing method that performs well on non-stationary signal processing in recent years,and can avoid the disadvantages of traditional non-stationary signal processing methods.In view of the achievements of EMD in one-dimensional non-stationary signal processing,researchers have tried to extend EMD into the field of images.And then Bi-dimensional empirical mode decomposition is developed,achieving good results in image processing such as image fusion and feature extraction.However,in practical applications,some problems have been found to be improved in the BEMD algorithm.In this thesis,the existing BEMD algorithm is studied and discussed from the theoretical and practical aspects.The main work is as follows:1.Aiming at the boundary effect,an adaptive variable length boundary extension method is proposed based on image self-similarity.The extension part gained by this method refers to the structural texture features of the image itself.It can,in theory,better reflect the changing trend of the image and the error caused by the boundary effect can be reduced effectively.Regarding the extension length of the boundary,this paper proposes a method for determining the length based on the extreme point distribution.The extension length can be adaptively determined according to the extreme point distribution characteristics of the data without artificial setting.The error introduced by the self-similar extension is minimized while ensuring that the extension part includes the extreme point.2.About the outer stop condition,the decomposition stop condition of BEMD is studied from the perspective of judging whether the dominant component is noise or effective information in intrinsic mode function: if the dominant component in intrinsic mode function is changed from noise to signal effective information,the decomposition stops.On how to determine what is the main component in the intrinsic mode function,this paper,after analyzing the shortcomings of the existing judgment methods,proposes a stop condition judgment criterion suitable for BEMD.It is proved by experiments that the judgment criterion proposed in this paper can accurately judge whether the high SNR or the lower SNR is used,so that the decomposition stops in time.3.The improved BEMD algorithm is used for image denoising,and a BEMD-based interval threshold denoising method is proposed.The interval threshold denoising method analyzes the characteristics of the BEMD decomposition result,and it regard the set of specific pixel points in the decomposition result as the threshold processing unit.The pixel values in the same unit are uniformly processed to achieve the purpose of image denoising.Compared with the existing BEMD-based denoising method and the traditional filtering denoising algorithm,it is proved that the improved method proposed in this paper has good performance in removing additive noise and multiplicative noise.