Panorama Stitching Algorithm Based On Statistics Of Similarity Of Texture

Virtual reality, multi-media and computer network are to be called the modern three greatest computer techniques. Virtual reality technique with three "I" , that is Immersion, Interaction and Imagination as its most essential specialty, combined with other advanced techniques will be widely used in future.Virtual reality based on graphic is constructed by means of computer graphic technique. First we abstract the real world to construct the 3D geometrical model by polygons, then after the site and direction of observation is specified, we can create a virtual reality scene after a process of polygons drawing, staining, concealing, lighting, projecting. The worst of all disadvantages of this technique is we have to invest manpower and material resources to create a mass of geometrical model. Expression engine restricted the complexity of scenes and quality of expression because of the real time property of roaming. So we should use special expression engine which limited uses of this technique.Otherwise in virtual reality based on images we can create a virtual space by this way: first we shoot pictures at the same site but in different directions, then stitch these pictures to get cylindrical or spherical surface panoramas after they were processed some way, last establish the relationships of these panoramas. In this virtual space the roamer can look around and move from one site to another smoothly. The scenes are vivid and we do not need construct mass of models in virtual reality based on graphic because of the using of pictures. This virtual reality system is not strict to the capability of computer and can be carried out in PC.One of important task in virtual reality based on images is stitching pictures to get panoramas. We need shoot a large number of pictures and stitch them after they were projected to create multi-site virtual scenes. Stitching pictures to get panoramas is an important taskbecause the quality of stitching result has deciding influence to the effect of roaming. There is plenty of stitching algorithm by far, but they have some common disadvantages such as bad precision, not able to overcome some disturbing factors such as the difference of lighting between pictures, rotary error happening when shooting. Aiming at resolving these problems this paper introduced a brand new stitching algorithm based on statistics of similarity of texture. It adopted a special matching algorithm that is statistics of similarity of texture. This matching algorithm adopts crisscross characteristic region. By analyzing information of texture in characteristic region it can overcome disturbing factors such as the difference of lighting between pictures, definition, variety of curtaining relationship, and rotary error. In this stitching algorithm, the operator should select two characteristic regions in one of the two pictures that are to be stitched. Then the algorithm will search the two best-matching region in the other picture, compute the angle of the rotary between the two pictures, rotate one of the two pictures, then stitching the picture, thereby the rotary error is corrected.In addition, this paper has introduced a criterion for matching algorithm. This criterion is valuable to designing, estimating matching algorithm.