Research And Implementation Of Key Technologies For Heterogeneous Image Registration And Fusion

With the rapid development of sensor technology,various sensors are widely used to collect variety of images.And the technology of collecting images by sensors is adopted in many fields.However,images collected by a single sensor in the same scene cannot express complete scene information in every time.In order to show more comprehensive scene information in one image,it is necessary to register and fuse images obtained by different types of sensors.This thesis mainly studies the key technologies and applications of heterogeneous image fusion.Among them,the technologies of heterogeneous image fusion include heterogeneous image registration and image fusion at the pixel level.The proposed image fusion method is implemented on the platform of DSP and GPU in real time.This thesis introduces two heterogeneous image registration methods based on SIFT features and mutual information respectively,and analyzes the advantages and disadvantages of the above two methods in real cases.Combining the advantages of the two methods in practical applications,the proposed method achieves the more accurate semi-automated heterogeneous image registration technology by introducing the manual assistance.This thesis focuses on the heterogeneous image fusion methods.At first,researches are conducted on the principles and the applications of robust principal component analysis(RPCA),wavelet transform(DWT)and non-downsampled contourlet transform(NSCT)in the field of heterogeneous image fusion.On this basis,this thesis proposes a double-transformation hybrid image fusion method based on RPCA,NSCT and DWT,and improves the fusion strategy according to the fusion framework in the proposed method.The double-transformation hybrid fusion method can effectively combine the advantages of RPCA,NSCT and DWT,and improves the fusion quality of the heterogeneous image fusion algorithm.Finally,this thesis compares the proposed method with the traditional methods by experiments.The experimental results show that the proposed method performs well in the subjective and objective evaluation of the fusion images.In the end,the real-time implementation of the image fusion methods is studied in this thesis.The fusion methods are implemented based on the platform of DSP and GPU respectively.Through the of code optimization,algorithm optimization and parallel computing,the real-time implementation is achieved in the fusion of infrared images and visual images,which enhances the practical value of these algorithms and expands their application fields.