|
|
|
| A Visual Mechanism Inspired Model for Image Segmentation |
| Key Lab for Neuroinformation of Ministry of Education, University of Electronic Science and Technology of China, Chengdu 610054, China |
|
|
|
Abstract In this paper, a multiscale Markov random field (MRF) model in the wavelet domain was proposed by simulating several image segmentation functions of the visual system. Human visual system (HVS) has feature detection ability, hierarchy, bidirectional connection, and self-learning mechanisms. For an input scene, our model provided its sparse representations using wavelet transforms (WT) to mimic the feature detection ability, and used pyramid framework to mimic hierarchy. In the framework of the model, there were two information flows that were used to mimic bidirectional connection, i.e., a bottom-up procedure to extract input features and a top-down procedure to provide feedback controls. In addition, the iteration in procession was the simulation of selflearning mechanisms, and the multiscale MRF was the tool for image segmentation. The quality of the framework was tested and compared to some classic image segmentation algorithms. Results showed that the proposed model obtained improved data than those obtained by classic image segmentation algorithms.
|
|
|
|
|
|
|
|
|
