Plural Figure Recognition on Frosted Glass Image Using Neural Networks.

This research is concerned with a new technology on image recognition of objects using neural networks. The standard method of pattern recognition is geometrical pattern matching. However, it is generally difficult to recognize figures on frosted glass because the brightness difference is pretty little. In this research, a new image recognition system is proposed by the application of image processing and neural networks. First, the figure pattern has been extracted from the frosted glass by using two kinds of smoothing filters, and transferred to a binary pattern by threshold value. Second, the extracted pattern including plural figures has been separated in single ones respectively through the technique using the labeling processing. This method is based on a combination of the labeling and the center of gravity. Third, a figure pattern with noise and broken form has been transformed into a well-regulated one by pattern exchange system using neural network. Finally, a figure pattern has been divided by neural networks, and the recognition result has been presented. Through experiments, the validity of this method is verified.