This paper studies the performance of binary image processing CNN templates when the actual template values at each cell are allowed to vary from their nominal values. We examine the validity of one plausible measure of the robustness to random template variations: the minimum absolute value of the current into the capacitor taken over all possible binary state patterns divided by the norm of the template elements. While this measure can be proven to be a valid indicator of robustness for linear threshold templates, its predictive power on the more dynamically complex CCD template is mixed. In some cases, an estimate of the error rate based upon this measure matches remarkably well with the results of numerical simulations. In others, this measure of robustness predicts that one template is more robust than another, while numerical simulations indicate that the opposite is true.
|Number of pages||6|
|Publication status||Published - Dec 1 1994|
|Event||Proceedings of the 3rd IEEE International Workshop on Cellular Neural Networks and their Applications (CNNA-94) - Rome, Italy|
Duration: Dec 18 1994 → Dec 21 1994
|Other||Proceedings of the 3rd IEEE International Workshop on Cellular Neural Networks and their Applications (CNNA-94)|
|Period||12/18/94 → 12/21/94|
ASJC Scopus subject areas