Research Of Image Fusion Method Based On Compressed Sensing Theory

Compressed sensing receive widespread attention because of its sampling theorem farbelow the Nyquist.Make use of the signal’s sparsity or compressibility, a lot of redundant datawas abandoned in the sample, and then restore the original signal by reconstruction algorithm.This theory can greatly reduce the amount of data of the signal acquisition, it is propitious to thesignal processing especially on the signal with a large amount of data. In this paper, the theory isapplied to image fusion, especially for the successful used in remote sensing images,offord a newmethod for image fusion, and provide technical support for the further development of the theoryof compressed sensing.Based on compressed sensing theory,the image fusion rules acting on compressed sensingdomain, the amount of data is only about30%of the original data, this method for the largeamount of data on remote sensing image fusion, will have a good time and economic benefits.Based on compressive sensing image fusion method, this paper do the following work:1. Image Sparse treatment. Using Fourier transform,Discrete Cosine transform and DiscreteWavelet transform image sparse process, Fourier transform or discrete cosine transform of theimage converted from the spatial domain to the frequency domain, discrete wavelet transformsource image decomposed into low-frequency portion and the high frequency part of theimage.after that most of the coefficient is0or close to0and concentrated in a particularregion.The three sparse approach can meet the compressed sensing theory of image sparsity.2. Measurement matrix design. After Fast Fourier transform,image information is mainlyconcentrated in the center of the image area, double stars shape measurement matrix is design toensure that most of the effective information can be collected.Similarly,according to the DiscreteCosine transform characteristic design ray-shaped measurement matrix,Discrete Wavelettransform design Dual-ray measurement matrix.Each measurement matrix sampling rate is onlyabout30%of the original data.3. Image fusion of compressed sensing domain. We use the maximum method, theminimum value method, the mean method and the weighting method four fusionmethod. Maximum value of the data may be characteristic information, maximum method isadopted to ensure the feature information obtained fusion. but the maximum value is alsopossible a noise point, in order to reduce the influence of noise,choice of the minimum valuemethod; The average method take the average value of data, taking into account thecharacteristics and suppression noise information; However, the above three fusion method arenot taken the size of the CS data information amount into consideration, this paper presents a method of fusion as a weighting coefficient according to the size of the amount of CS datainformation-the weighting method.4. Comparative and analyse the fusion results.This paper uses three kinds of the sparse baseand corresponding measurement matrix,four fusion method to dill with the10sets of images,atotal of120kinds of fusion results were acquire. From the experimental results, the fusion resultsobtained by the different sparse and fusion method is obvious differences. In the same sparseway the weighting method fusion results in each group is relatively more excellent; In samefusion method, the Fourier sparse fusion superior.5.Compare with the traditional fusion methods.Traditional image fusion method chosenwavelet fusion method and IHS fusion method. In comparison with the IHS fusion method, thepanchromatic and multispectral fusion result by IHS is much better;but in the medical images,optics images and SAR image fusion is superior to the traditional wavelet fusion results.