Abstract
Purpose: :
In practice, prosthetic vision may be distorted by the nonlinear mapping between the stimuli and the perception in the visual field, which would influence object recognition. In this study, we evaluate the object categorization performances by retinal prosthetics through simulated distorted vision
Methods: :
Ten subjects performed object categorization tasks with simulated distorted vision triggered by 8×8 and 16×16 channel implants, respectively. The distorted visual stimulations, e.g. pixelized images of spoon, tea pot, automobile and human face, were created by either shifting the upper half of the visual field from 1/8 to 8/8 of the field width, i.e., translation distortion, or altering the curvature of the visual field from 0.002 to 0.016, i.e., barrel distortion. Categorization accuracy (CA) was measured as the performances of object recognition.
Results: :
Overall CA decreased under distorted stimulations (e.g., translation distortion: 63.9%±4.1% and barrel distortion: 67.0%±3.3%) compared to that under undistorted images (75.0%±3.9%). Paired t-test demonstrated that distortions did significantly influence the automobile recognition (p<0.05), but not interfere with the face recognition (p>0.05). Further analysis of different distortion levels in object categorization indicated that (1) CA was proportional to the shifts of translation distortion and (2) barrel distortion didn’t decrease CA unless the curvature was more than 0.006. Moreover, the averaged CA was 57.9%±4.8% for 8×8 channel prosthetics and 78.1%±2.5% for 16×16 channel prosthetics under distorted stimulation. To obtain a critical CA value (CAt=62.5%) with the presence of distortions, we suggested that at least a 10×10 prosthetics should be implanted according to linear regression model.
Conclusions: :
In order to recognizing a distorted prosthetic vision, at least 10×10 channel prosthetics should be implanted. It was also suggested from our results that prominent contours of objects which are robust to distortions should be used for prosthetic vision to have better object categorization.
Keywords: face perception • low vision • receptive fields