Purchase this article with an account.
Koji Nozato, Qiang Yang, Kenichi Saito, Jie Zhang, David R Williams, Ethan A Rossi; Automated correction of sinusoidal distortion and drift in resonant scanning retinal imaging systems. Invest. Ophthalmol. Vis. Sci. 2014;55(13):1599.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
In many raster scanning systems, such as an adaptive optics scanning light ophthalmoscope (AOSLO), the imaging field is formed by a slow scanner and a fast resonant scanner. The resonant scan motion is sinusoidal rather than linear, causing a spatial distortion in the image. Typically, a grating is used to calibrate the distortion, but this is time-consuming. Moreover, the resonant scanner output that reports the scanner position in time drifts relative to the true position due to mechanical and electrical instability, causing image drift. Both problems are difficult to correct without additional hardware that directly monitors the mirror. We use the retinal image itself for this purpose, obviating the need for both calibration and additional hardware.
In Fig. 1, the synchronization signal (H-sync) drifts such that the latency, T1, is variable. The offset between the retinal images obtained on the forward and backward scans (acquired during the intervals shown as rectangles in Fig. 1) provides the information needed to lock the image sampling windows at the optimum time relative to the true mirror position. We use cross-correlation to calculate this offset, which is then divided by two and used to dynamically adjust T2 so that the sum of T1 and T2 are constant. By choosing the correct constant, the image sampling windows are fixed at the center of the most linear portion of the sinusoidal motion. T2 was updated every 3-5 seconds. The sinusoidal distortion in the image is then linearized by inverting the sinusoidal relationship between space and time.
We verified our method by imaging 10 human eyes. Before applying our method, the forward and backward images were typically misaligned by 4.9% of the image size (~4.4 arcmin in a 1.5° imaging field) and this increased to 8.9% after 2 hours. After alignment, the two images were precisely centered on each linear half of the sinusoidal curve, with 0.17% accuracy (~9.4 arc seconds), and the sinusoidal distortion was removed. This method is robust even with low contrast images.
This method accurately corrects sinusoidal distortion and drift without additional hardware, facilitating image interlacing and efficient use of data acquired from both scan directions. This approach, which eliminates the need for manual calibration, is a step toward the ultimate goal of a turnkey AOSLO.
This PDF is available to Subscribers Only