Research On Multi-sensor Information Fusion Technology Based On Visual Saliency

With the continuous development of optical imaging technology and digital information processing technology,the scene information reappeared by a unitary sensor can no longer meet the requirements of optical target detection in complex scenes.Scholars and technicians begin to combine various types of sensors.Multi-sensor information fusion technology has been widely used in military,security,medicine and other fields.Multi-source image fusion is one of the typical applications of multi-sensor information fusion.Images in the same scene are acquired by highly complementary image sensors of different imaging types and the images are then fused for further applications.Visual saliency represents the ability of targets to attract the attention of observers in a particular scene.Introducing this feature into image fusion can extract targets more accurately.At present,fusion algorithms may often lose detail information or be unable to highlight important targets when processing multi-source images.Therefore,it is particularly necessary to carry out in-depth research on multi-sensor information fusion technology.Multi-sensor information fusion based on visual saliency is researched in this paper,mainly focusing on multi-source image fusion of infrared image and visible image.The specific work is as follows:In order to study how to improve multi-sensor information fusion technology,the basic theory of image fusion is researched from the aspects of image fusion level,basic methods of pixel-level image fusion and image preprocessing.Also,the basic theory of visual salience is researched from the aspects of visual salience overview,common visual features and visual salience model.Therefore,visual salience is applied to image fusion to synthesize the information of multi-source images,which is more favorable for subsequent target detection and other processing.The subjective and objective image quality assessment methods are also studied as important evaluation references for multi-sensor information fusion algorithms.Aiming at the problem that infrared image details are blurred and visible light images are difficult to highlight thermal targets,an image fusion algorithm based on visual saliency and guided filtering is designed in this paper.Non-local mean filtering is used to decompose the image into two scales.An improved frequency tuned algorithm based on bilateral filtering is used to detect the visual saliency map,and iterative processing is carried out to eliminate pseudo saliency targets in the infrared image.Low frequency information fusion is carried out by combining the matching score with the visual saliency map optimized by guided filtering,and high frequency information fusion is carried out by combining Laplacian energy with the visual saliency map.This algorithm improves the problem of losing detail information of visible images and better highlights the significant targets of infrared images.Aiming at the problem that multi-source image fusion algorithms can only process one kind of multi-source images,another image fusion algorithm based on visual saliency and nonsubsample contourlet transform(NSCT)is designed.The decomposition method based on NSCT is utilized to decompose the source images into multi scales.And the improved visual saliency algorithm is utilized to extract saliency maps from source images respectively.The low frequency part of the images is fused by using the matching score between saliency maps.The high frequency part is fused by comparing the 8-neighborhood Laplacian energy.This algorithm not only performs well in infrared and visible image fusion,but also has advantages in multifocus image fusion.