Remote Sensing Image Registration Using Fast Visual Saliency and Improved SIFT

Image registration is the matching process between two or more images at different times and by various sensors, in different imaging conditions and view angles. Scale-invariant feature transform (SIFT) is the benchmark algorithm for this registration. However, the algorithm is computationally expensive. In this article, a saliency detection method was proposed. The saliency detection method identifies the target area, and key points are extracted by SIFT algorithm in this area. The extraction of key points only in the target area causes the algorithm be fast as the valuable key points are extracted. The algorithm accuracy is suitable in comparison with the competing methods which makes the proposed saliency detection algorithm be a suitable choice in real-time applications. The employed descriptor is gradient local orientation histogram (GLOH). Finally, the pruning of the correct key points was performed by fast sample consensus (FSC) algorithm. Experiment results showed that the proposed method increased the computational efficiency by about 10% compared to other methods.