Pre-processing, registration and selection of adaptive optics corrected retinal images

Area of Science:

• Ophthalmology
• Biomedical Imaging
• Image Processing

Background:

• Adaptive optics (AO) systems provide high-resolution retinal imaging.
• Processing sequences of fundus images presents challenges in illumination and registration.

Purpose of the Study:

• To enhance processing of AO flood illumination fundus images.
• To correct uneven retinal illumination.
• To automatically select high-quality images and precisely register them.

Main Methods:

• Pre-processing retinal images using wavelet-based illumination correction.
• Evaluating image quality with metrics like sharpness, variance, and contrast.
• Two-stage image registration (coarse cross-correlation, fine parabolic interpolation/maximum-likelihood estimation) with rotation correction.

Main Results:

• Wavelet approach (Daubechies 4) effectively corrects illumination, enhancing sharpness and contrast.
• A 'Designer metric' accurately assesses image quality, correlating well with visual evaluation.
• Sub-pixel registration accuracy achieved (RMS difference 0.047 pixels); rotation correction significantly improves image edges.

Conclusions:

• Wavelet-based pre-processing is effective for AO-assisted cone photoreceptor imaging.
• The 'Designer Metric' provides reliable image quality assessment.
• Two-stage registration with rotation correction substantially enhances final image contrast and sharpness.