Based On Information Distribution Composite Feature For Multi-modal Image Registration Research

The development of remote sensing sensor pushes forward the field of remote sensing research.Conventional single-mode images are increasingly unable to adapt to the current remote sensing technology due to the limitation of monotonous imaging band.Since the 21 st century,the application of multi-source remote sensing information has begun to have sufficient data support.The comprehensive application of multi-source remote sensing data has been widely concerned by many scholars.At present,multi-source remote sensing data has been comprehensively applied in geological disaster monitoring,terrain monitoring,national defense engineering and other fields.Multi-modal image registration is the first step in the comprehensive application of multimodal image.The accuracy of multi-modal image registration directly affects the accuracy of subsequent steps.Due to the difference of imaging principle and imaging band,there are strong geometric distortion and nonlinear radiation differences among multi-modal images.And there are various types of noise inevitably between multi-modal images.These problems make multimodal image registration based on feature face the following difficulties: feature extraction with high similarity;Reasonable and detailed descriptor construction;Point of interest extraction with high repetition rate.In response to challenges in the field of multi-modal image registration,this paper makes some contributions,specifically as follows:(1)To solve the problem of difficulty in extracting features with high similarity,we propose adaptive information entropy graph(AIEM).From the perspective of information theory,the method classifies the image signals adaptively and calculates the information content in the neighborhood pixel by pixel.Because the nonlinear radiation difference will only change the intensity of different types of signals nonlinearly,it will not affect the result of signal classification.Therefore,AIEM is robust to nonlinear radiation differences,and is externally manifested as image texture features.The experiment shows that the AIEM of multi-modal images has high similarity.(2)The detailed and perfect feature descriptor is constructed.Because the information distribution features of multi-modal images have high similarity,a composite feature description model is proposed.The composite feature description model consists of maximum information index and information trend.The maximum information index describes the main changing direction of image information.The information trend describes the transformation mode of the overall information of the image.These two single descriptors are highly complementary in capturing changes,so the compound feature description model can fully describe the characteristics of information changes.(3)FAST detector was used to extract interest points on AIEM.Experimental results show that,under the same parameter Settings,AIEM has the highest repetition rate and number of points of interest.This greatly improves the registration efficiency of the whole algorithm.The experiment shows that the extracted points of interest have high repetition rate.