Abstract
Purpose: :
Current numerical vision aids optimize the presented image by globally modulating its features, such as size, luminance, color and contrasts. We developed a novel, real-time video processing which identifies the moving pixels of the image, highlights he moving objects or persons, and thus improves figure-ground discrimination.
Methods: :
We designed a parallel optic flow algorithm coupled with a parallel rendering algorithm that specifically enhances parts of the image that are moving. That processing was tested binocularly in nine individuals (62 to 90 yo, mean 71,2 yo) with bilateral atrophic AMD (binocular visual acuity 0,06 to 2 /10, mean 0,012). The procedure was as follows. Ten video sequences (7-13 sec in duration) were recorded. They feature a room where, at a random time, 1 or 2 persons enter into the viewing field. Recordings were shown to tested subjects, both in original and processed versions, 3 times each. All videos were presented in a random order on a 21 inch monitor, at a 50 cm distance. Subjects indicated when they detected moving features. For each recording, detection latency was compared between original and processed versions.
Results: :
Detection latency of moving features was shorter in all tested individuals using processed recordings, compared to the original one. In five of the nine subjects, latency reduction was significant (paired t-test: p=0.0014 to p=0.0001). Globally the difference was also significant (Wilcoxon paired test on all subjects: p=0.002)
Conclusions: :
This method introduces an ecological hierarchy for feature enhancement. It presents to the patient more salient features that are better perceived and interpreted. It improves figure ground segregation, and allows affected patients to detect quicker a moving person. In the future, this algorithm will be embedded in digital glasses for augmented vision to directly alleviate visual handicap.
Keywords: image processing • aging: visual performance • clinical (human) or epidemiologic studies: systems/equipment/techniques