SAR And Visible Images Fusion Based On Structural Sparse Representation Of Multi-scale

As an important subfield of image processing,image fusion has a very important status and value of research.Due to the restriction of research technology and cost,various sensors have their own disadvantages and fail to always obtain high quality images.Image fusion technology can unite the complementary information of two or more images from disparate sensors in the same scene organically through a certain algorithm.The fusion image will contain more abundant and dependable information.Synthetic Aperture Radar(SAR)is widely used in military reconnaissance and national economy because of its all-time,all-sky and penetrating characteristics.Since the SAR is based on the coherence theory,coherent speckle noise will be formed in the SAR image.Speckle noise will have a certain impact on the resolution and recognition of the target in the SAR image.At the same time,the SAR image is a gray image.It loses the spectral information of the target,which is not good for the human eye observation.Optical sensors can collect spectral information of objects.However,it is easy for optical sensors to disturb by external factors such as weather.By merging the SAR with the visible image,an image with rich details and spectral information can be obtained.Although the traditional fusion method based on sparse representation(SR)can reduce the noise to a certain degree,it is mainly aimed at additive noise and can not fundamentally solve the noise problem in the fusion image.Besides,the fused image obtained by the traditional SAR and color visible image fusion method has serious spectral distortion,which is not conducive to the subsequent processing of the human eye and machines.In view of the above problems,this article has presented the fusion method of SAR and visible image based on structural sparse representation of multi-scale.The work of this paper mainly includes the following two points:(1)Aiming at the problem of loss of detail information caused by sparse representation of image blocks with different structural features using the same over-complete dictionary,this paper proposes a SAR and visible image fusion algorithm based on adaptive structural sparse representation.Firstly,a series of sub-dictionaries which can accurately represent various structural features need to be constructed through a high-quality image data set.Then,appropriate sub-dictionaries will be selected according to the respective structural features of the image blocks and sparse representation is carried out.The corresponding fusion rules are formulated according to the activity level and sparsity degree of the sparsity coefficients.Finally,the final fusion image can be obtained by multiplying the sparsity coefficients obtained by the corresponding sub-dictionaries.Through the analysis and discussion of the experimental results,it is proved that the algorithm can effectively eliminate the coherent speckle noise in the fusion image while retaining more detailed structure information.(2)The traditional fusion image based on HSI transformation has rock-ribbed spectral distortion,which is mainly affected by the high frequency part in the luminance component.In order to ensure that the detail structure information is not lost while reducing its spectral distortion,this paper proposes a SAR and color visible image fusion method based on multi-scale SR and spectral correction.Firstly,this method decomposes the image into low-rank and sparse components by robust principal component analysis(RPCA).Then,the fusion rules based on sparse representation and multi-scale anisotropy are adopted for low-rank and sparse components,respectively.The initial fusion luminance component is obtained.Subsequently,spectral correction is performed according to the structural features of the initial fusion luminance component.Finally,the final color fusion image can be obtained by HSI inverse transformation.Through the simulation experiment and analysis of 5 groups of measured data,compared with the common fusion algorithm based on multi-scale and sparse representation,it is proved that the algorithm is more optimal in structure retention and spectral distortion.