Abstract
Purpose: :
Watson and Ahumada (2008 [http://journalofvision.org/8/4/17]) proposed an image-based simulation model for predicting acuity as a function of optical aberrations. In this model a "neural" image is computed from the optics and a neural contrast sensitivity function. White 'internal' noise is added to the image and it is cross-correlated with internal templates. The response to the image is then that of the highest correlating template. In the uncertainty-free version of the model, the performance is only a function of the matrix of cross-correlations between the neural images and the templates, the matrix of cross-correlations among the templates, and the internal noise level. They presented metrics derived under the assumption that there is no correlation among the templates. We derived and tested a metric based on the assumption that the correlations among the templates are caused by a single factor. This metric looks like their independence-assumption metric with the noise level for each difference reduced by the correlation between the two templates.
Methods: :
Model predictions were computed for blurred and unblurred tumbling E images and Sloan letters. Model simulations were done with 10,000 trials in each condition using Watson and Ahumada's fast simulation method.
Results: :
For the tumbling E's, the single factor model predictions of the average per cent correct were extremely accurate. For the Sloan letters, the average per cent predictions are good, but the metric does not correctly predict the relative accuracy among the different letters. Predictions improve as the blur increases.
Conclusions: :
The fast simulation results provide the full confusion matrix.When only the proportion correct is needed, the single-factor metric provides a useful approximation. The acuity model can be regarded as a general model for image recognition. The usefulness of the single-factor approximation will in general decline as the number of image templates increases.
Keywords: visual acuity • aberrations • pattern vision