Feature Based Automatic Multiview Image Registration

Automatic image registration is a vital yet challenging task, particularly for remote sensing images. A fully automatic registration approach which is accurate, robust, and fast is required. The registration process is divided into six main steps: feature detection, feature extraction, feature matching, outlier detection and removal, transform model estimation and resampling. In the feature detection, keypoints (PCA-SIFT), blob (SURF), region (MSER) features are detected. The features are then matched to produce point correspondences. The alignment process uses the RANSAC algorithm to estimate an outlier free transformation using point correspondences. The target image is transformed by using transformation function, which results in automatic image registration. Mutual information is used for fine tuning registration. In this thesis, feature descriptors used are SURF (Speeded Up Robust Features), MSER (Maximally Stable Extremal Regions) and PCA-SIFT. They are invariant to zoom, noise, scale, rotation and illumination, hence very useful. The analysis is conducted for various image transformations such as image rotation, scaling and change in illumination. For all these transformations, various quality checking parameters such as recall and RMSE error are evaluated to analyse the performance of registration. In the registration of multiview images, the affine transformation model applied is not suitable. So a more appropriate transformation models such as thin-plate spline is subtitled the affine model in the proposed work.