Research And Implementation Of Infrared Omnidirectional Image Mosaic Algorithm System

Recently,in the field of infrared imaging,it is impossible to obtain an image with high spatial resolution and large field of view,due to the fact that the size of the infrared arrays is small.Therefore,there has been large amount of large-field imaging methods based on multi-frame image mosaic.However,due to brightness difference,environment,and low registration accuracy,obvious stitching trace occurs in the mosaic infrared images,such as bright dark mutations or structural misalignment around the edges in the overlapping region.Besides,when the target moves in the overlapped area,the phenomenon named ghost appears,which seriously reduces the performance of image mosaic methods.Traditional infrared image mosaic algorithms only deal with a single problem mentioned above,i.e.,only eliminating the stitching seam or handling the ghost problem.Moreover,target pixels in the overlapped part will be segmented,and the integrity of targets be destroyed,reducing the quality of outputs.In order to improve the quality of stitching,this thesis presented an algorithm via detecting moving target area for infrared image stitching.First,obtaining a optimal stitching seam without segmenting the target pixels,and then we adopt an improved fade-in and fade-out fusion method to stitch the images.Finally,an infrared mosaic image without obvious problems of stitching trace and ghost is obtained.This method consists of detection of the moving target area,generation of optimal stitching seam,and image fusion method.In this paper,we utilize the morphological dilation method to deal with the detection results of the inter-frame difference method,which can be finely divided via the background difference method.And then,the target location is estimated.Experimental results show that the presented method achieve a good result in dealing with problems of hole and double shadow in the inter-frame difference method,and enhance the robustness of the background difference method.Subsequently,to eliminate the ghost phenomenon while targets move in the overlapped area,it is suitable to find an optimal stitching seam,via which obtains the smallest difference between the two images.However,since the target has relative motion between the two images,most traditional methods based on searching for the best stitching seam tend to cut target pixels,and even destroy the integrity of the target.Therefore,based onthe idea of dynamic programming,we introduce the moving target detection technology,i.e.,to extract the detected moving target area as an unsearchable area.Moreover,the presented method avoids the target area by extending the original search direction and finally obtains an optimal stitching seam with whole target pixels.The experimental simulation results describe that the presented method can effectively avoid the appearance of ghost and the target pixels to be cut.In order to improve the performance of eliminate the stitching trace in stitching images,image fusion of the stitched image is realized after obtaining the optimal stitching seam.Nevertheless,most traditional fusion methods focus on the overlapped area,resulting in generating the ghosting problem and a low processing efficiency.Hence,in this paper,we studied a method based on improving the fade-in and fade-out method.First,the average gray value difference between the two sides of the optimal is calculated,and a suitable threshold is set.In addition,the weight coefficients are redistributed according to the detection results,in case of the loss of detail information in the original images.Finally,experimental results demonstrate that this method achieve a better stitching performance in either eliminating the geometric stitching seam,handling the ghosting problem or improving the efficiency.