Image noise can dramatically affect image processing and hemoglobin oxygen saturation (SO<sub>2</sub>) calculation accuracy in non-invasive retinal oximetry. Recently, the denoising algorithm based on Variance stabilizing transform (VST) and dual domain filter (DDID) has been proposed to address this issue by our lab. Actually, dual-wavelength retinal images belong to multi-mode images, in order to maximize the use of complementary information between the dual-wavelength images, we further improve the algorithm. Firstly, noise parameters were also estimated by mixed Poisson-Gaussian (MPG) noise model. Secondly, a novel MPG denoising algorithm which we called VST+CDDID was proposed based on VST and cross dual domain filter. To evaluate the proposed algorithm, both simulative and real experiments have been carried out and the results show that the proposed method can effectively remove MPG noise and preserve edge details. Compared with current denoising methods based on single mode, such as VST+DDID and VST+ block-matching 3D filtering (BM3D), the proposed method shows great advantage in terms of visual quality and low-contrast detail. In conclusion, VST+CDDID can effectively use the complementary information between multi-mode images and combine the advantages of both cross bilateral filter in the time domain and Short-Time Fourier Transform (STFT) in the frequency domain. And it effectively restrained ringing effect by alternating iterative.
Noninvasive measurement of hemoglobin oxygen saturation (SO2) in retinal vessels is based on spectrophotometry and spectral absorption characteristics of tissue. Retinal images at 570 and 600 nm are simultaneously captured by dual-wavelength retinal oximetry based on fundus camera. SO2 is finally measured after vessel segmentation, image registration, and calculation of optical density ratio of two images. However, image noise can dramatically affect subsequent image processing and SO2 calculation accuracy. The aforementioned problem remains to be addressed. The purpose of this study was to improve image quality and SO2 calculation accuracy by noise analysis and denoising algorithm for dual-wavelength images. First, noise parameters were estimated by mixed Poisson–Gaussian (MPG) noise model. Second, an MPG denoising algorithm which we called variance stabilizing transform (VST) + dual-domain image denoising (DDID) was proposed based on VST and improved dual-domain filter. The results show that VST + DDID is able to effectively remove MPG noise and preserve image edge details. VST + DDID is better than VST + block-matching and three-dimensional filtering, especially in preserving low-contrast details. The following simulation and analysis indicate that MPG noise in the retinal images can lead to erroneously low measurement for SO2, and the denoised images can provide more accurate grayscale values for retinal oximetry.
The dual-wavelength retinal image registration is one of the critical steps in the spectrophotometric measurements of
oxygen saturation in the retinal vasculature. The dual-wavelength images (570 nm and 600 nm) are simultaneously
captured by dual-wavelength retinal oximeter based on commercial fundus camera. The retinal oxygen saturation is
finally measured after vessel segmentation, image registration and calculation of optical density ratio of the two images.
Because the dual-wavelength images are acquired from different optical path, it is necessary to go through image
registration before they are used to analyze the oxygen saturation. This paper presents a new approach to
dual-wavelength retinal image registration based on vessel segmentation and optic disc detection. Firstly, the multi-scale
segmentation algorithm based on the Hessian matrix is used to realize vessel segmentation. Secondly, after optic disc is
detected by convergence index filter and the center of the optic disc is obtained by centriod algorithm, the translational
difference between the images can be determined. The center of the optic disc is used as the center of rotation, and the
registration based on mutual information can be achieved using contour and gray information of vessels through
segmented image. So the rotational difference between the images can be determined too. The result shows that the
algorithm can provide an accurate registration for the dual-wavelength retinal image.
To better understand how the eye’s optics affects stereopsis, we measured stereoacuity before and after higher-order aberration (HOA) correction with a binocular adaptive optics visual simulator. The HOAs were corrected either binocularly or monocularly in the better eye (the eye with better contrast sensitivity). A two-line stereo pattern served as the visual stimulus. Stereo thresholds at different viewing durations were obtained with the psychophysical method of constant stimuli. Binocular HOA correction led to significant improvement in stereoacuity. However, better eye HOA correction could bring either a bad degradation or a slight improvement in stereoacuity. As viewing duration increased, the stereo benefit approached the level of 1.0 for both binocular and better eye correction, suggesting that long viewing durations might weaken the effects of the eye’s optical quality on stereopsis.
A novel adaptive optics vision simulator (AOVS) is presented and characterized for several design features, including automated measuring and compensating eye’s aberrations up to the fifth order, which fully cover aberrations typically found in the human eye, even for the cases of highly aberrated eyes. Especially, it is equipped with 35 elements bimorph deformable mirror with bigger stroke and smaller size, which could help establish near-diffraction-limited ocular optics condition. To investigate the validity of this apparatus, pilot data under different aberration correction pattern from one subjects are collected, and contrast sensitivity function (CSF), an important psychophysical function in vision, is obtained also. Results from living eyes show a practically perfect aberration correction and demonstrate the utility of this system.