An automated SAR image registration approach using hidden Markov scale Invariant Feature Transform algorithm

Image registration process ensures the geometric and locational accuracy of the images and provides geometric registration between multi-temporal image set and/or vector based dataset. Manual registration procedure is time expensive and receptive to user based errors such as ground control point marking error. Considering the main drawbacks of the manual process, this study aims to investigate the performance of improved Scale Invariant Feature Transform (SIFT) algorithm in automated image registration process. SIFT algorithm relies on features each of which is invariant to image scaling and rotation and robust to local geometric distortion. In application, one ortho-rectified satellite image with high positional accuracy is selected as reference. Feature vectors are extracted from this reference image in order to be used as training feature dataset. Then object recognition and location transformation is applied on the images at different dates/sensors belonging to same geographic area. The efficiency of algorithm is first verified by using optic sensor data (SPOT 5-6) then applied to the SAR data (RSAT 2).