多尺度模型下的自动图像变形

传统变形技术以交互式特征对应为基础,建立源图像到形变图像的映射函数。而交互特征对应是一个耗时费力又缺乏精确度的过程,因而自动特征对应是一项十分有意义的研究。论文在总结经典变形算法及已有的自动变形算法基础上,提出了在多尺度空间中搜索形变函数的变形算法,将自动变形问题转化为能量模型下的优化问题。算法在满足精度要求的同时,快速地逼近全局最优形变函数,达到自动特征指定的目的。实验表明,算法能够在较短的时间内得到令人满意的插值序列。

Automatic morphing with multiscale analysis

Traditional image morphing techniques involve user interaction,which is usually boring and inaccurate.Therefore,research on automatic feature-correspondence is of great significance.Based on classic morphing algorithms and their automations, this paper presents an automatic morphing algorithm,which exploits Multi-Resolution Analysis.It turns the feature-correspondence problem to an energy-minimization process,and experimentation demonstrates that,optimum warping function can be found in a short time,and satisfactory in-between images come out.