An Image Fusion Method Of Atrous Algorithm Based On Non-separable Wavelet

In recent years,the multi sensor image fusion technique has received wide attention in machine vision,remote sensing,medical,military and other fields.The multi sensor image fusion refers to the images about the same object or scene of multiple sensors to collect,carry out comprehensive treatment according to a certain algorithm,to generate a new image.This image cannot get by a single sensor.There is both redundancy and complementary exist between the multi sensors images,so multi sensor image fusion technique can improve the reliability of the system but also can improve the utilization efficiency of image information.Currently,Multi-resolution analysis methods as tensor order wavelet or non-separable wavelet transform has been widely used in image fusion.But the tensor order wavelet and the non-separable wavelet also has its limitations.The tensor order wavelet is a two-dimensional wavelet generated by two one-dimensional wavelets.Because of the one-dimensional wavelet species restrictions,the tensor order wavelets species is very terminate.In image fusion process,the tensor order wavelet can only obtain information from both horizontal and vertical directions.The amount of information it gets is limited.The non-separable wavelet can make up for this deficiency.The non-separable wavelet is not subject to the constraints of the tensor product.Just design a set of filters to meet certain conditions;you can get the corresponding non-separable wavelet.Moreover,the non-separable wavelet can obtain information in multiple directions.So its image fusion results can have a high spatial resolution.However,its spectral information has great deviation compared with the original image.Furthermore,the algorithm also has a great amount of calculation.By studying atrous wavelet theory and methods,we propose a new image fusion method for atrous algorithm based on non-separable wavelet.The algorithm combines the atrous wavelet who can maintain good spectral information.And the algorithm process does not require sampling and interpolation.We use the low-pass filter of the non-separable wavelet in place the convolution operators of the atrous wavelet in image fusion.In this article we study the non-separable wavelet basic theory and methods of two-channel and four-channel.The non-separable wavelet can maintain high spatial resolution in image fusion.But it is inadequate to maintain the spectral information.So the algorithm based on the translation invariance of atrous wavelet,can keep good texture.We put the algorithm for image fusion,the fusion results both in spatial resolution or in the spectral information have very good results.In addition,the image fusion method proposed in this paper also inherited the simple and easy features of the atrous wavelet.And the time and space required to run the algorithm is small.So it is easy to program.