Research On Image Processing Methods Based On Novel BEMD

The traditional bi-dimensional empirical mode decomposition(BEMD)can decompose an image into several intrinsic mode function(IMF)components with different scale oscillation modes according to a sifting processing,which has been applied to many fields of image processing and analysis.However,such methods extract IMFs by calculating envelope surfaces with the help of interpolation and sifting processing.They still face many difficulties and challenges,including over-/under-shootings in the interpolation,inefficiency for images with large sizes,and deficiency in the convergence guarantee.In order to solve these problems,some novel BEMD methods are proposed,but they still have some defects in multi-scale feature description,salient edge protection,applications in image processing and other aspects.In order to improve the results of the existing BEMD methods and the application ability of BEMD in the field of image processing and analysis,this thesis makes some studies on several novel BEMD methods and their applications in image enhancement,image fusion,image watermarking,etc.The main work of this thesis can be summarized as follows:1.We propose a novel BEMD method based on scale-guided optimization.It first obtains an initial IMF of the original image according to the unconstrained optimization model based on Delaunay triangulation,and then improves the initial IMF by a scaleguided optimization model.The scale-guided optimization model uses the length of the initial IMF as the scale guide,and relaxes the constraints near each local extremum point of the initial IMF by a smoothing operator to eliminate the redundant oscillations.If an edge-aware smoothing operator is used in the optimization process,BEMD and edge-aware smoothing can be integrated into a unified framework.The experimental results show that the proposed method not only can clearly represent the features of different spatial scales of the input image,but also can perform edge-aware smoothing for the input image.Therefore it can retain the salient edge features of the original image which often are lost in the existing BEMD methods during decomposition.2.We propose a low-light color image enhancement method based on scale-guided optimization BEMD and Retinex.It firstly makes full use of the scale-guided optimization BEMD proposed in this paper,which can clearly represent the different spatial scale features of the input image and can perform edge-aware smoothing.Secondly,it carries out single-scale Retinex on the different scale components obtained from the decomposition of the bright channel of the input color image to restore the detail information under the shadow.Finally,the Retinex enhancement results of different scales are combined,and the final enhancement image is generated by the gray balance and color restoration operation.The experimental results show that the proposed method can demonstrate the detail and color characteristics of the original images well.Comparing with some famous multi-scale Retinex methods,the proposed method not only can reduce the artificial effects near the strong edges of the input image without selecting the smooth scale manually,but also has some advantages in terms of visualization and objective evaluation.3.We propose an image fusion method via morphological filter based BEMD and feature guidance.It firstly decomposes the input images into several IMF components and a residual quickly according to the multi-channel BEMD based on morphological filter.Secondly,the components obtained from BEMD are fused using the overlapping partition block fusion strategy to reduce the noises generated by the pixel-by-pixel fusion strategy used in the existing BEMD-based image fusion methods.In order to obtain the fusion results with much higher quality,it adopts an energy-based maximum selection rule to fuse the IMF components,and uses the feature information extracted from the IMF components as a guide to fuse the residual.The experimental results show that the proposed method has obvious advantages over the existing image fusion methods based on BEMD in terms of visual effect,objective evaluation and time performance.Furthermore,it is also very competitive with the other type image fusion methods.4.We propose a color image multiple watermarking method via morphological filter based BEMD and cyclic embedding.It firstly decomposes the color host image according to the multi-channel BEMD based on morphological filter.And then,it selects the positions of watermark embedding by detecting the extrema of the decomposed IMF components.After that,it can obtain 1D watermark signals by dimensionality reduction of multiple watermark images,which can be repeatedly embedded into the three RGB channels to increase the robustness of the proposed watermark method.The embedding strength is calculated by presetting the lower limit of the peak signal-to-noise ratio(PSNR)of the embedded watermark image to ensure the invisibility of the embedded watermark.The final watermark is obtained by statistical voting on multiple cyclic embedded watermark information of each channel image without using the original host image.The experimental results show that the proposed method can almost extract the watermark completely under some common attacks when the host image with watermark information has a big PSNR value.Compared with several color image watermarking methods developed in recent years,the proposed method not only has better invisibility,but also obtains good results in resisting attacks.