Purchase this article with an account.
S. Dua, R.W. Beuerman; Image Stabilization for Real-time Clinical Confocal Microscopy . Invest. Ophthalmol. Vis. Sci. 2003;44(13):3658.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Purpose: Examining real-time images of the eye is made more difficult due to intrinsic movements from factors such as activity of the extra-ocular muscles, heart beat and the inability of the subject to maintain a steady gaze. The problem of detecting and compensating for these movements is complex because the change can be attributed to one of any combination of the following variations: rotation, translation, and z-axis translation. Methods: Although, the operator can adjust for large scale movements of a patient, minute changes (or destabilization) will remain undetected due to their low amplitude and frequency. However, at a typical magnification of about X500, these movements make appreciation of displayed images difficult. To detect and correctly report the error vector from these destabilized images is the objective of this research. In the proposed approach, the images registered by the confocal microscope are re-routed to algorithmic analysis software that segments the images, extracts descriptors, identifies areas of destabilization and then stabilizes the images prior to storage on recording media. A suitable error correction vector is also calculated to be routed back to the confocal microscope for automated positional correction. A unique indexing schema called FG-index has been developed, employed and computational benefits have been found. Results: Real-time images are collected from confocal and frame extracted for analysis. Analysis is performed on images with translation of +-300, rotations up to 11 degrees and various permutations of both. The proposed techniques can detect similar frames with a peak accuracy of 96% (for 28 samples). Areas of destabilization in an image frame are also identified and displayed for studies of anomalous trends. Distributed error correction vectors are computed and images then "locally" corrected at areas of destabilization. The corrected frames are then amalgamated to be recorded on the storage media. A "global" error-vector is also computed and supplied to the confocal for fast image correction of subsequent frames. Conclusions: The destabilized images collected from confocal experiments have local errors rather than global errors (over entire image) as believed earlier. These local errors can be algorithmically computed and corrected for stabilization. We conclude our discussion with rigorous experimental results on real-data, and we suggest that this corrected data stream will make image evaluation quicker and easier.
This PDF is available to Subscribers Only