基于角点匹配和纹理映射的前向序列图像插值算法

Image Interpolation of Foreword Moving Image Sequence Based on Texture Mapping

在基于图像的绘制(Image—based Rendering简写为IBR)中,有效快速地实现两幅图像间的平滑过渡是一个关键而困难的问题。本文提出一种基于角点匹配和纹理映射的方法,完成对前向拍摄有较大尺度和较小旋转变化的图像进行插值。该算法先检测角点并通过Delaunay三角化匹配角点,再利用匹配点的凸包分图像为中心和边缘区域两部分,用不同的方法建立两个区域和原图像的对应网格关系,最后用一个快速三角形纹理映射的方法建立插值图像。对实际图像试验证明该方法可以快速,有效和实时地产生高质量的插值图像。

In image-based rendering（IBR）applications, how to effectively and efficiently interpolate two consecutive images is a crucial and difficult problem to rend smooth transition. This paper presents a new and efficient algorithm to synthesize two consecutive images with relatively big scale change but little rotation change. First, the algorithm detects the corners and uses the Delaunay triangulation to match the corners. According to the convex hull of the matched corners, the algorithm divides the images into the center and the border regions in which different methods are employed to map pixels between the two images. Then, a fast texture - mapping method is used to map pixels in the triangular regions. Experimental results conducted on real images demonstrated that our method could produce superior image - based rendering effect in real-time.