Research On Image Fusion Technology Based On Multi-scale Analysis

Image fusion,as a discipline spanning many fields of science and technology,has been attract attentions of various circles of society both at home and abroad in recent years.Whether it is military night operations,radar warning,target survey,or civilian video surveillance,medical equipment and other aspects all play a pivotal role.With the rise and development of sensor manufacturing industry,there are many kinds of image sensors.By using a variety of sensors to collect information and images of the target scene,it is helpful to improve the deficiency of single sensor and make people understand and recognize better to the target scene.With the development of image fusion,the field of image fusion based on multi-scale analysis is gradually becoming the research focus of scholars at home and abroad,and fusion algorithms spring up.Through consulting a large number of the scholars of literature at home and abroad,this thesis collates and analyzes the principle of image algorithm based on multi-scale analysis.Then,this thesis puts forward two improved algorithmsand verifies its practicability under the embedded platform with Raspberry Pi as the core.The main contents of this thesis are as follows:(1)The process,level,fusion rules and evaluation methods of image fusion theory are studied in detail,which lays a foundation for the improvement and analysis of subsequent algorithms.(2)The classical algorithms in the field of image fusion based on multi-scale analysis are deeply studied.The simulation experiments of Contourlet algorithm and NSCT algorithm are carried out to understand the specific steps and advantages and disadvantages of the algorithm.(3)This thesispresents two kinds of image fusion algorithms based on the domain of NSCT.One is a dynamic weighted fusion algorithm based on regional energy.The algorithm calculates the regional energy of low-frequencysub-band in the lowfrequency part of NSCT.The final low-frequencysub-band fusion image is determined by dynamic weighting.The other is the dynamic WNMF algorithm based on NSCT domain,which uses the dynamic WNMF fusion rule to preserve the detail information of the low-frequency part effectively by using the abrupt change degree as the threshold in the low-frequency part.The fusion rule based on the regional energy matching degree is used in the high-frequency part,the absolute value is taken for the highest level of high-frequency,and the area square is used for the other layers.Choose the absolute value.In order to judge the fusion effect of the two algorithms,two groups of infrared and visible images under strict registration and a group of multi-focus images are selected for the fusion simulation experiment.Through subjec-tive and objective evaluation criteria,the fusion algorithm with relatively high fusion effect is obtained.(4)The embedded image fusion system based on Raspberry Pi is constructed.The dynamic WNMF algorithm based on NSCT domain is selected and two groups of infrared and visible images which are strictly registered are selected.And a group of camera images which are registered by SIFT algorithm are fused into the hardware system.According to the evaluation criteria of subjective and objective fusion,the hardware platform is proved to be practical.