基于结构张量的图像插值方法

Image interpolation method based on structure tensor

提出了一种新的图像插值算法,该算法利用局部结构张量所描述的图像几何特征增强了图像的边缘而不会产生伪影。在仿真实验中,应用该方法能够得到比传统的双线性和双三次插值方法更优的结果,特别是在边缘区域。而且该方法采用的插值格式能有效地减小计算量,适合实时应用。就提出的插值模型和一种基于变分的插值方法之间的关系进行了讨论,分析表明后者只是该模型的一个特例。

An efficient method for image interpolation was presented, in which geometric features described by the local structure tensor were exploited to enhance the sharpness of edges without incurring any additional artifacts. Simulation results show that the proposed approach exhibited better performance than the standard linear interpolation, particularly in the edge regions. The implementation has low complexity and it is well suited for real-time applications. In addition, the relationship between the method and an interpolation scheme based on variational models presented in previous work was also discussed and further analysis shows that the latter is a special case of our algorithm.