Research On Image Fusion Algorithms Using Bidimensional Empirical Mode Decomposition

Image fusion is a procedure to produce a new fused image from the multisource images.The fused image is not only clearer and more understandable,but also decipherable for human perception.As an adaptive multi-scale image analysis method,BEMD can decompose the image into a series of BIMFs and a residual.Image fusion algorithm based on BEMD is studied in this paper.The main contents related to the study are as follows:(1)Image fusion theory is studied,introducing the process of image fusion,image fusion level and image quality evaluation methods.According to the degree of information abstraction in the image fusion process,the image fusion level is divided into pixel-level?feature-level and decisionlevel image fusion.The image fusion method based on BEMD belongs to the pixel-level image fusion.(2)EMD theory is studied.EMD is a multi-scale analysis method for non-linear and nonstationary signals.BEMD is the bidimensional extension of EMD,which performs multi-scale analysis on images to acquire the complex information in the image.(3)Fusion method of the infrared and visible images based on BEMD and NMF is proposed.Image fusion between the infrared and visible images can enhance vision and improve the situation awareness.The visible image is transformed into IHS components.The intensity component and the infrared image are decomposed into high frequency components and the low frequency components by means of BEMD.The NMF fusion rule and the arithmetic mean rule are respectively applied to fuse the high frequency components and the low frequency components.The new intensity is transformed back into RGB.The fused image based on the proposed method has the significant object information and sharp edge information.(4)Fusion method of the infrared and visible images based on improved BEMD and the window energy correlation fusion rule is proposed.A direct use of the bidimensional empirical mode decomposition(BEMD)method for image fusion suffers from a high computation cost.Instead of the time-consuming surface interpolation methods,the improved BEMD adopts the order statistics filter?the limited neighborhood method and the smooth filter to calculate the envelope of the surface,,so as to accelerate the sifting process within the original BEMD.Based on the characteristic of the improved BEMD,the improved BEMD is proposed and applied to fuse of high frequency components.The fused image based on the proposed method is not only able to achieve the best objective evaluation metrics,but also consistent with human visual perception.(5)Fusion method of the infrared and visible images based on fast BEMD and deep learning is proposed.The fast BEMD algorithm adopts only one order statistics filter to formulate the mean envelope,which accelerates the algorithm.The image fusion rule based on deep learning is proposed.The fusion rule calculates the fusion weights according to the multi-layer features extracted by the convolutional neural network.Experimental results demonstrate that the proposed method exhibits the best fusion performance and outperforms some state-of-the-art methods and representative in terms of both subjective visual quality and objective evaluation metrics.