Research On Hierarchical Fusion And Stitching For3D Image Reconstructed By Image Sequences

Stitching technique for3D reconstruction of image sequences is one of thedirections of the three-dimensional reconstruction. Research for3D Stitchingtechnique is conductive to better achieve the3D reconstruction of image sequences,which has important academic value and broad application prospects. Currently, mostresearches on image fusion and stitching are focused on the plane image at home andabroad, although the techniques and methods of which is not directly used for thefusion and stitching of three-dimensional image, but the plane image fusion andmosaic methods could play a certain reference for3D image fusion and stitching. Thispaper does the research on layer fusion and stitching method for three-dimensionalimages reconstructed by image sequences. The primary coverage is as follows:1. Access to many literatures at home and abroad about stereo matching andimage fusion and mosaic, and introduce the study situation of this technique.Elaborate the related theory of stereo matching and image fusion and stitching,including the basic knowledge of stereo registration, the classification of stereomatching algorithms, the processes of image stitching, the techniques of image fusion,etc.2. In the research and analysis of the current stereo matching method based onthe area of the window, this paper proposes a matching method based on improvedregional window。First, we divide the color image into different parts and compute thecost of each segment; second, we compute the cross adaptive window for matchingpixel and obtain the cost of cross adaptive window; third, we combine the twomatching cost into the final matching cost. In addition, in order to improve theefficiency of the parallax match, this paper also adopts an improved parallax rapidsearch strategy, narrowing the scope of parallax search. Comparing with the originalalgorithm, the accuracy of matching is improved in the texture area.3. Based on the studying of the theory of multi-scale decomposition of the imageas well as the method of multi-scale image fusion, it proposes a method of the imagefusion method based on NSCT transform. NSCT transform is used for the multi-scaleand multi-direction decomposition of two images, then the high frequency componentis selected through the rule of regional contrast energy and the low frequencycomponent by adopting the combination of local gradient energy and the weightedaverage coefficient selection method, finally experimental results show this fusionmethod is better than other multi-scale fusion method.4. The two images which have parallax and different views are stitching bymeans of traditional mosaic method that will appear ghost phenomenon. Based thisreason, this paper proposes a hierarchical fusion and stitching method based on thedepth information. Firstly, the images are divided into different layers by the depthinformation; secondly, the image of the different layers is fused by different fusion methods respectively; finally, the fused background is covered by the fusedforeground and the better fusion and stitching image is obtained.