Image Recognition using Morphological Operators on Cellular Neural Networks
| Project/Area Number |
14550406
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
System engineering
|
|---|
| Research Institution | Takuma National College of Technology |
|---|
Principal Investigator |
SUMITOMO Kazuhiro Takuma National College of Technology, Department of Electronics, Professor, 電子工学科, 教授 (20044688)
|
|---|
| Project Period (FY) |
2002 – 2004
|
| Project Status |
Completed (Fiscal Year 2004)
|
| Budget Amount *help |
¥1,300,000 (Direct Cost: ¥1,300,000)
Fiscal Year 2004: ¥500,000 (Direct Cost: ¥500,000)
Fiscal Year 2003: ¥300,000 (Direct Cost: ¥300,000)
Fiscal Year 2002: ¥500,000 (Direct Cost: ¥500,000)
|
|---|
| Keywords | Cellular Neural Networks / Morphology operators / Neural Networks / Learning method for image recognition / リアルタイム画像認識 |
|---|
| Research Abstract |
Image recognition in real time with a compact system will have vast applications. If we use Cellular Neural Networks (CNN) for an image recognition system, we could have a compact system with real time processing. The main purpose of this research is to develop a basic method of image recognition utilizing CNN.
Cellular neural networks (CNN) have two prominent features, i.e., the high speed data processing ability by their parallel and analog circuits, and the feasibility to be built on an 1C by their restricted local connections between cells. However, their most shortcoming point is the difficulty to determine the connecting weights between cells, i.e., the difficulty of programming comparing with the conventional processors. For the simple image processing such as edge detection, hole-filling, etc., the author has developed a general designing method of templates using fuzzy inference. For more complicated applications such as image recognitions, we need to devise a new templates designing method. To reduce the difficulty of designing templates, I have engaged with morphological image operators. Since the morphological image operators have been developed as a powerful tool to analyze shapes, we had better utilize them on CNN. As the first result of this research, I have shown that morphological image operators can be implemented on Discrete Time Cellular Neural Networks (DTCNN) and some Japanese characters are recognized with Hit-or-miss operator. As the second result of this research, a learning algorithm for the construction of the structuring elements is proposed using neural networks equivalent to Hit-or-miss operation in DTCNN. With the learned structuring elements, the simulation results show that even characters of different size and of different rotation angle can be successfully recognized.
|
Report
(4 results)
Research Products
(10 results)
