- Collection:
- Atlanta University and Clark Atlanta University Theses and Dissertations
- Title:
- Hybrid neural network image processing testbed, 1996
- Creator:
- Hall, Kevin D.
- Date of Original:
- 1996-07-01
- Subject:
- Degrees, Academic
Dissertations, Academic - Location:
- United States, Georgia, Fulton County, Atlanta, 33.749, -84.38798
- Medium:
- theses
- Type:
- Text
- Format:
- application/pdf
- Description:
- The focus of this research was to establish a testbed for pattern recognition In this testbed, the wavelet transform is used as a preprocessor for various neural networks. The wavelet transform is used to perform image compression, and several wavelet filters and compression techniques are implemented The compressed data is later formatted and used as input to a neural network where pattern recognition is performed. The wavelet filters used in the wavelet transformation were the Daubechies 4 (DAUD4) and the Haar wavelet filters. After compression was performed, the root mean square error (RMS) was computed and compared with a 'common' compression technique called JPEG compression. After testing each compression technique, zone compression using the wavelet transform yielded the best results. At this point, the compressed data was used by various neural networks for pattern recognition. There were three neural nets in the testbed They were the neocognitron, a genetic algorithm driven neural network, and the Hopfield neural net. Each neural net was used to perform pattern recognition using the compressed data The results from each neural net were good, but the neocognitron gave the best results.
- External Identifiers:
- Metadata URL:
- http://hdl.handle.net/20.500.12322/cau.td:1996_hall_kevin_d.pdf
- Rights Holder:
- Clark Atlanta University
- Holding Institution:
- Atlanta University Center Robert W. Woodruff Library
- Rights:
-
