Research On Image Fusion Technology For Panorama Stitching

Panoramic stitching technology is an important research interest which has a great position in the field of digital image application. The technique provides composite image with a wide field of view, high resolution and rich information content which has an advantage of overall understanding and comprehensive recognition of target scene. It is widely applied in the field of military, medicine and virtual reality system, as well as the need of people’s daily entertainment and personal interest.As the last step of stitching process the effect of image fusion algorithm designed plays an important role in the performance of image fusion. Image fusion technique has been developing for years, resulting in a set of complete and developed theory. However, some adjustment and improvement is need for new application background.A new algorithm process of image fusion is designed for panoramic stitching, in the paper, to resolve the problem which stitching image with image sequence captured by hand-held device may emerge, such as fuzzy boundaries, ghosting and exposure difference. Paper tries to search for the technology intersection and application background in common by means of studying many technologies in digital image process. Moreover, Boss of algorithm processes of image stitching and each module are analysis, as well as the characteristics of the source image acquired by handheld terminals and specific application scenarios. A new fusion algorithm is proposed. Firstly, region segmentation is achieved by edge detection. And then make image fusion by calculating each split areas.The main contents are as follows:Proposed a colour image edge detection algorithm based on feature points, which extracted edge uses similar manner of region growing. Firstly, a gradient edge detector and an colour similarity descriptors are defined. Seed points are generated via image feature points and got grow by the mean of breadth traversal. Both thresholds to judge new edge points and criteria of growth stopping are designed. Finally, the true edge is got by recall method. This edge extraction method has a high accuracy and a single pixel width response. Meanwhile, the character of continuity makes image segmentation realized, which can achieve a finer level of the region segmentation, to provide the conditions for the subsequent algorithm.Proposed a region-based image fusion algorithm. First boundary tracing algorithm is improved to realize regional identity and orderly sequence of points. Second cross matrix of overlapping images is defined to merge regions. Last gradient field is designed and image sequences are stitching using Poisson fusion method. The resulting fused image maintains clarity of image details well and provides a smooth transition between image sequences.Image fusion algorithm arise in paper can make the brightness variations due to differences between the image exposure well smoothed. Meanwhile there appears to have a great performance in suppressing ghosting effects, eliminating blurred edges. The final result meets with observation consistent coherence of the human eyes.