Canon Kabushiki Kaisha

Andrew Docherty

Peter Alleine Fletcher

James Austin Besley

Barry James Drake

Ruimin Pan

Tatiana Dudina

A displacement field relating positions of pixels between at least two images having an initial displacement field is adjusted. Each pixel of the first image and the second image is associated with a label. A first pixel in the first image maps to a second pixel in the second image based on a vector of the initial displacement field. A set of third pixels includes the second pixel and a plurality of adjacent pixels to the second pixel, the third pixels being associated with labels and a candidate shift vector with respect to the second pixel. A covariance score is determined for each of the third pixels, the covariance score defining a statistical dependence between the first label of the first pixel and each of the third labels. An adjustment shift vector is then determined based on the covariance scores and the candidate shift vectors of the third pixels.

Image registration method

A method determining a retinal pigment epithelium identifies a plurality of regions using captured image data of the eye and fits a curve into at least some of the regions. A curve score is determined associated with the fitted curve using at least a distance between the fitted curve and at least some of the regions in which a contribution of the regions to the curve score is biased towards (asymmetric) regions below the fitted curve. These steps are repeated whereupon one of the fitted curves is selected, using the corresponding associated curve score, for classifying some of the regions as forming at least a part of a retinal pigment epithelium.

Robust segmentation of retinal pigment epithelium layer

A method of compensating for distortion, in a plurality of retinal image frames, arising from eye movement receives at least one reference image frame and plural other image frames of the eye. For each frame, the method determines a plurality of shifts relative to the reference image frame, and an initial eye movement trajectory based on at least an initial reference frame distortion, and the shifts within each of said other image frames. The method estimates a current reference frame distortion using the initial trajectory and the shifts within each of the other image frames, and a current eye movement trajectory based on the estimated current reference frame distortion, and the shifts within each of said other image frames to provide a smooth transition between the other image frames in the eye movement. With this the method compensates for distortion in said frames using the current eye movement trajectory.

Registration across frame boundaries in AO-SLO capture

A computer-implementable method of analyzing tissues of a retina. A polarization-sensitive (PS-OCT) image data set of the retina is received from a polarization-sensitive device, each element in the polarization-sensitive image data set being associated with an intensity value ({I, . . . , I}) and a polarization direction ({P, . . . , P}). A likelihood score is determined for each element of the polarization-sensitive image data set based on the intensity value of the element and a degree of divergence (&#x3bc;*P or D) of the polarization direction associated with the element from a reference polarization direction (&#x3bc;*) associated with the polarization sensitive image data set, wherein the likelihood score indicates whether said element is drawn from a directionally polarized distribution or a depolarizing distribution, Elements of the polarization-sensitive image data set are classified using the determined likelihood scores to analyze tissues of the retina.

Depolarizing region identification in the retina

A method of determining a coordinate transform between a first image and a second image, said method comprising the steps of: determining a rate of change of pixel values for locations on the first image to identify candidate alignment patches in the first image; specifying subsets of patches from the set of candidate alignment patches based on an error metric, selecting a subset of candidate alignment patches from said plurality of subsets of candidate alignment patches based upon a predetermined criterion; estimating, for each patch in the selected subset, a shift between the patch and a corresponding patch in the second image, the location of the corresponding patch in the second image being determined from the location of the patch in the first image; and determining the coordinate transform between the first image and second image based on the estimated shifts.

Growing community of inventors

St. Leonards, Australia

Andrew Docherty