A generic framework for image rectification using multiple types of feature

In photogrammetry, the traditional image matching and precise rectification is mainly based on point features, which are simple, intuitional and accurate. In many cases, however, it is difficult to acquire accurate ground control points in the areas where cross points and corners are not available, thus the point-based precise rectification is unfeasible. On the other hand, features such as straight lines, free-form curves and areal regions are usually more stable than point-based features and can be utilized to cope with the problem of missing points and to register image accurately. In this paper, a generic framework for image precise rectification using multiple features, including points, straight line segments, free-form curves and areal regions is proposed. Firstly, a generic framework for image rectification using multiple features is established based on the generalized distance, which differs for different types of features. Secondly, a robust and smooth Hausdorff distance is proposed for curve-based and area-based geometric correction. The continuity and derivability of the novel Hausdorff distance makes it possible to minimize the distances via gradient descent approaches. Thirdly, the generalized distance is specified by the existing point-based and straight line-based distances and the suggested curve-based and area-based distance. Finally, uniform error equations are integrated into the geometric correction models based on multiple features. The experimental results show that the generic framework is reliable for image rectification, and can be applied in multi-source images (SAR and optical image).