Research On Feature Extraction And Fusion Of Infrared And Visible Images In Complex Background

With the development of optoelectronic technology,optical imaging is becoming more and more mature and popular in the infrared and visible bands.Visible imaging technology can capture specific information of the scene,so visible images tend to have rich appearance information and high spatial resolution.However,under th e influence of weak light or bad weather environment such as haze,sandstorm,and smog,the imaging effect of visible sensor is poor,which is not beneficial to the recognition of hot target.Infrared imaging systems use the difference in temperature or ra diation between the background and hot target to generate an infrared image.Infrared image can reflect the thermal radiation of the object.The higher the temperature of the object is,the more prominent the scene information description is.However,comp ared with visible images,infrared images usually suffer from a series of problems such as low resolution,lack of details,and poor clarity.In view of this,To obtain more comprehensive and effective information,we can combine multiple images acquired by infrared and visible sensors to make full use of the complementary characteristics of source images.However,most image fusion algorithms tend to lose edge,texture,thermal information and other features in the process of feature extraction,especially the complex background of infrared and visible image fusion is hard to realize that visible image details or infrared thermal radiation information is easy to be lost in the process of feature extractionTo solve the above problems,this thesis proposes a feature extraction and fusion method for infrared and visible images under complex background,which mainly solves the loss of detail texture information and thermal information in fused images.The main contents of the method are as follows:(1)To enhance the ability to express the background details of the fused image,this thesis proposes a fusion method of infrared and visible images based on deep wavelet-dense network(WT-Dense Net).WT-Dense Net consists of wavelet block and dense block,which are used to extract low frequency feature and deep feature of source image respectively.The proposed network makes full use of the advantages of wavelet transform to separate high and low-frequency features and deep density network to extract deep features,and then achieves multi-feature aggregation by weighted addition strategy.Compared with single feature extraction block,multi-feature extraction makes fusion image contain more details.Because of the increase of low frequency characteristics,the fusion image performance is smoother,the visual effect is better,and the image quality is higher.(2)To preserve as much thermal information as possible while enhancing detail information in the fusion process,this thesis proposes a high-quality infrared and visible image fusion method based on iterative differential thermal information filter(IDTIF).Our method is basically divided into two steps: Firstly,the dynamic threshold thermal information filter(DTTIF)is used to adjust the pixel values of the thermal target and the background area,and the thermal target enhancement is achieved by expanding the relative pixel difference between the thermal target and the background area;secondly,a feature fusion method of multiple difference rolling guidance filter feature fusion method(DRGFM)is designed to realize feature enhancement of visible images.The proposed methods and other classical fusion methods are tested on multiple sets of image experimental datasets with complex backgrounds,the experimental results show that our methods not only improve the expression ability of detail background in fusion results,but also retain thermal targets to a great extent.It can be seen from the subjective and objective evaluation indicators that our algorithms show better fusion capability than other algorithms.