quality assurance systems
test images recognition
This article discusses the use of machine vision systems for quality control of information display facilities. Results of simulation of test images (alphabets) recognition processes by neural network algorithms and optimal filtering algorithm to exposure to noise are shown. At research were simulated two and three-layer neural networks with a fixed structure.
The test alphabet consisted of ten symbols in size 5×5 elements. Networks training was spent till the moment of reception of a total error between individual criterion function and real-world signals at the network output at consecutive recognition of all ten symbols not exceeding 0.001.
From the simulation results follows that the quality of recognition algorithms neuronetwork noisy test images close to theoretically possible and slightly concede to the optimal filtering algorithm. It's more than three layers increases do not lead to significant increases recognition quality. Increase in quantity of knots in the intermediate layers more number of receptors of the first layer slightly improves recognition characteristics.
Research findings confirm the prospect of neuronetwork algorithms on systems to evaluate the quality of information display facilities.