Deviating Angular Feature for Image Recognition System Using the Improved Neural Network Classifier.
Date of Award
Doctor of Philosophy (PhD)
J. Bush Jones
The ability to recognize images makes it possible to abstractly conceptualize the world. Many in the field of machine learning have attempted to invent an image recognition system with the recognition capabilities of a human. This dissertation presents a method of modifications to existent image recognition systems, which greatly improves the efficiency of existing data imaging methods. This modification, the Deviating Angular Feature (DAF), has two obvious applications: (1) the recognition of handwritten numerals; and (2) the automatic identification of aircraft. Modifications of feature extraction and classification processes of current image recognition systems can leads to the systemic enhancement of data imaging. This research proposes a customized blend of image curvature extraction algorithms and the neural network classifiers trained by the Epoch Gradual Increase in Accuracy (EGIA) training algorithm. Using the DAF, the recognition of handwritten numerals and the automatic identification of aircraft have been improved. According to the preliminary results, the recognition system achieved an accuracy rate of 98.7% when applied to handwritten numeral recognition. When applied to automatic aircraft identification, the system achieved a 100% rate of recognition. The novel design of the prototype is quite flexible; thus, the system is easy to maintain, modify, and distribute.
Srinakarin, Adisak, "Deviating Angular Feature for Image Recognition System Using the Improved Neural Network Classifier." (1998). LSU Historical Dissertations and Theses. 6870.