Registration method for multispectral retinal images

摘要

A method for performing registration of multispectral retinal images. Corresponding cross-over points and bifurcation points of blood vessel in an eye are identified in a fixed image and in an offset image. The relative displacement of each point, between the images, is calculated. Based on these relative displacements, the offset of each pixel in the offset image is interpolated and the offset image is transformed into a corrected (distorted) image. Such an image can be used to assess the health of the eye. Further, a fixed image and a offset image of an eye are divided in a tile pattern. The relative displacement of each tile of the offset image with respect to the corresponding tile in the fixed image is calculated. An offset for each tile is calculated and the offset image is corrected as a function of those offsets.

主权项

1. A method to register retinal images, the method comprising:
dividing a first retinal image acquired at a first optical wavelength into first image portions; dividing a second retinal image acquired at a second optical wavelength into second image portions, each second image portion having a corresponding first image portion; determining a deformation vector for each second image portion with respect to its corresponding first image portion; identifying second image portions that have a deformation vector that fails a pre-determined criteria, to obtain identified second image portions; calculating, for each identified second image portion, an interpolated deformation vector in accordance with deformation vectors of neighbour second image portions; substituting, for each identified second image portion, its deformation vector with its corresponding interpolated deformation vector; and registering second image portions to their corresponding first image portions in accordance with deformation vectors that pass the pre-determined criteria and in accordance the interpolated deformation vectors.