An algorithm for the fast registration of image sequences obtained with a scanning laser ophthalmoscope

There is a need to quantify images of retinal pathology so that the natural progression and response to various forms of treatment such as laser therapy can be reliably monitored. The images obtained with a scanning laser ophthalmoscope are better than those from a traditional fundus camera, and the method of data collection leads naturally to digitization and quantitative techniques. However, during any angiographic sequence, there will inevitably be eye movement and it is essential that this be corrected for prior to the application of quantitative analysis. This article presents an algorithm (the smart localized binary correlation algorithm) for the automatic registration of fluorescein or indocyanine green angiograms which is based on localized correlation of binary templates in binary thresholded gradient images. The use of this algorithm decreases the calculation time up to 180 times when compared with the fastest applicable algorithm found in the literature. The algorithm only detects translational movement of the images but proves to be robust against rotational movement for the small angles (< 4 degrees) encountered in ophthalmic imaging. Registration accuracy is estimated to be +/- 2 pixels.