The Reaearch Of Pixel-level Multi-sensor Image Fusion Technology Based On Independent Component Analysis

In recent years，with the fast development of science and technology，multi-sensor imagefusion technology plays an important role in industrial field，medical field，military field andremote sensing field with its many advantages. Image fusion can be divided into pixel levelfusion，feature level fusion and decision level fusion，the paper mainly studies pixel levelimage fusion technology.The paper firstly introduces fundamental theory of image fusion，includes concepts，advantages，applications and levels of image fusion；Secondly discusses subjective andobjective evaluation methods of image fusion，puts forward a new image fusion objectiveevaluation method based on gradient mutual information；Then studies and compares someclassical pixel level image fusion algorithms，points out advantages and disadvantages ofvarious algorithms through fuses four group different sensors images and adopts entropy，standard deviation，mutual information，spatial frequency，gradient mutual information toevaluate the fused image，at the same time we also prove the effectiveness and feasibility ofgradient mutual information evaluation method；Finally image fusion based on independentcomponent analysis is introduced，we propose a new image fusion algorithm based onindependent component analysis，the basic idea of algorithm is trianing images from differentsensors separately and gets their mutual statistical independent basis function，then obtains themain area and the background area of source images，fuses the transformed ICA coefficientsthrough some fusion rules，reconstructs the fusion image，we reach a conclusion through thesimulation results and data analysis that our algorithm has better results than the classicalimage fusion algorithms.