2003 Fiscal Year Final Research Report Summary
Intelligent Preprocess Algorithm
| Project/Area Number |
12044212
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research on Priority Areas
|
| Allocation Type | Single-year Grants |
|---|
| Review Section |
Science and Engineering
|
|---|
| Research Institution | Kyusyu Institute of Technology |
|---|
Principal Investigator |
YAMAKAWA Takeshi Kyusyu Institute of Technology, Dept of Brain Science and Systems Engineering, Professor, 大学院・生命体工学研究科, 教授 (00005547)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
SAMATSU Takashi Kyusyu Tokai University, Dept. of Management Science, Assistant Professor, 応用情報学部, 講師 (60299667)
MIKIT Sutomu Kyusyu Institute of Technology, Dept of Brain Science and Systems Engineering, Associated Professor, 大学院・生命体工学研究科, 助教授 (20231607)
|
|---|
| Project Period (FY) |
2000 – 2002
|
| Keywords | Wavelet Neuron / Global Minimum / Local Stochastic Infomation / Stochastic Independency / Infomax ICA / Self-organizaing Maps / Code Book / Fuzzy Inference |
|---|
| Research Abstract |
The following research results have been obtained;
Filtering based on Wavelet Neuron and Local Statistical Information (2000) Noise reduction by low pass filtering and edge enhancement, which are complementary functions, are realized by the nonlinear description ability and the global minimum convergence of the Wavelet Neuron and verified with the real images. The hardware implementation of the system was achieved and exhibited the operating speed of 150 times as fast as software processing.
Blind Source Separation of Mixed Image and Sound Applicable to Any Kinds of Statistical Independency (2001) By fusion of Infomax ICA presented by Bell & Sejnowski with self-organizing maps, Blind Source Separation (BSS) of any kinds of statistical characteristics was implemented. It can be achieved by adequate nonlinear mapping of the output signals, the nonlinearity of which is obtained by self-organizing maps from input audio signals and input images.
Image Enlargement and Feature Extraction (2002) A pixel of the image is estimated by the surrounding pixels by two methods. The one employs the fuzzy inference and the other does the code book obtained by the self-organizing map. Both cases exhibited better quality of enlargement. The feature extraction was achieved by the wavelet network and implemented with FPGAs.
|
