Research of image registration algorithm based on template matching
Abstract
In order to provide support for image registration, the application of template matching in image registration is studied. This paper sums up four types, introduces the principle of each algorithm, and compares their advantages and disadvantages through simulation experiments: Based on the gray information of the algorithm is simple and real-time, but for complex image and low gray contrast images, cross matching results; mathematical transform based on the complex, large amount of calculation; mutual information does not require preprocessing, registration effect is good, is the research hotspot at present, but ignores the spatial relationship between pixels
Keywords
Download Options
Introduction
Image matching is one of the traditional research directions in image processing. In practical application, It is often required by different sensors at different time, register the images of the same scene in the space position under different imaging conditions, or find the corresponding target in another picture according to the known template. Therefore, image matching [1-2] is the basis of image processing techniques such as image rectification and image fusion. In recent years, image registration [3-5] research has sprung up, especially in medical image [6-9] research. From the theoretical and conceptual point of view, it is the development of image matching technology, and some of the methods and ideas in image matching still play a special role. From this point of view, this paper sorts out the related research of image registration from the perspective of template matching, and analyzes their respective characteristics in combination with simulation.
Conclusion
The results show that the algorithm based on gray-level information is simple and real-time, and it is sensitive to noise. The calculated amount is proportional to the size of the input image. For the image with complex image and low gray contrast, the matching result is cross- The Fourier-Mellin method is still correct for the picture after the rotation translation. The mutual information does not need to be preprocessed and the registration effect is good. It is the current research hotspot, but ignores the spatial position between the pixels relationship.