Name
Abstract | ||
|---|---|---|
We describe a hardware solution to a high-speed opticalcharacter recognition (OCR) problem. Noisy 15 × 10 binaryimages of machine written digits were processed and applied asinput to Intel's Electrically Trainable Analog Neural Network(ETANN). In software simulation, we trained an 80 × 54× 10 feedforward network using a modified version ofbackprop. We then downloaded the synaptic weights of the trainednetwork to ETANN and tweaked them to account for differencesbetween the simulation and the chip itself. The best recognitionerror rate was 0.9% in hardware with a 3.7% rejection rate on a1000-character test set. |
Year | DOI | Venue |
|---|---|---|
1993 | 10.1162/neco.1993.5.6.885 | Neural Computation |
Keywords | Field | DocType |
noisy digit,a1000-character test set,fast recognition,modified version ofbackprop,synaptic weight,software simulation,rejection rate,recognitionerror rate,high-speed opticalcharacter recognition,feedforward network,electrically trainable analog neural,hardware solution,neural network,error rate,optical character recognition,chip,binary image | Computer science,Word error rate,Binary image,Optical character recognition,Speech recognition,Software,Artificial neural network,Rejection rate,Feed forward,Test set | Journal |
Volume | Issue | ISSN |
5 | 6 | 0899-7667 |
Citations | PageRank | References |
2 | 1.13 | 0 |
Authors | ||
2 | ||
Name | Order | Citations | PageRank |
|---|---|---|---|
| Jeffrey N. Kidder | 1 | 2 | 1.13 |
| Daniel Seligson | 2 | 2 | 1.13 |