Global tracking of the ocular fundus pattern imaged by scanning laser ophthalmoscopy

This paper presents an algorithm for the automatic global tracking of ocular fundus landmarks in video image sequences generated by scanning laser ophthalmoscopy (SLO). The tracking algorithm is based on the computation of the discrete unnormalized cross-correlation of an interactively preselected small template and respective images in time sequences of fundus patterns. The correlation is executed with binary images derived from an automatical threshold limitation of the grey-value images. Due to the [+1/-1] representation of the binary images the unnormalized correlation functions directly relate to the Hamming distance of the template and the objects in the images. Experiments show that even templates with features distorted by noise are accurately recognized at any position. The accuracy of position detection is better than 0.4%. Possible hardware implementations of the algorithm which would reduce computation time are briefly mentioned.