Related Experiment Videos
Image recovery and segmentation using competitive learning in a layered network.
1Dept. of Comput. Sci., Univ. of Central Texas, Killeen, TX.
IEEE Transactions on Neural Networks
|January 1, 1996
Summary
This study introduces a novel iterative algorithm for image recovery and segmentation using competitive learning within Markov random fields (MRFs). The method enhances edge preservation and offers comparable boundary enhancement compared to existing techniques.
Area of Science:
- Computer Vision
- Image Processing
- Machine Learning
Background:
- Image recovery and segmentation are crucial in image processing.
- Markov random fields (MRFs) provide a probabilistic framework for image analysis.
- Existing methods like mean field approximation have limitations in edge preservation.
Purpose of the Study:
- To develop an iterative algorithm for image recovery and segmentation.
- To apply the principle of competitive learning within the MRF framework.
- To improve edge preservation and enhancement in image processing.
Main Methods:
- Formulated image recovery as an energy function minimization problem within MRFs.
- Developed a pixel-wise local update rule as a gradient descent for a global energy function.
- Investigated the relationship between the developed update rule and Kohonen's update rule.
Main Results:
- Introduced quantitative measures for edge preservation and enhancement.
- Demonstrated superior edge preservation compared to mean field approximation.
- Achieved comparable performance in boundary enhancement.
Conclusions:
- The proposed competitive learning-based iterative algorithm effectively addresses image recovery and segmentation.
- The algorithm shows significant improvements in edge preservation for noisy and real images.
- The method offers a viable alternative to existing MRF-based image processing techniques.