An Image Registration Method for Engineering Images

Abstract

In this paper, we proposed an image registration method for digitized cadastral images in Taiwan. The proposed method is composed of three parts: feature selection, RANSAC-based transform parameter estimation, and image stitching. In the feature selection, Harris Corner Detection is first used to extract corners as feature points for each engineering image and then some feature points are selected manually. To reduce the impact of wrong matched pairs on transform parameter estimation, the RANSAC-based transform parameter estimation is developed. After removing the wrong feature point pairs, the least squares error estimation method is used to estimate transform parameters. The image stitching between source image and reference image can be performed based on the estimated transform parameters. Experimental results show that the proposed method can not only effectively select suitable feature point pairs for parameter estimation but also stitch source image and reference image well.