Abstract
Most recent attempts at object segmentation have been based on object motion in video sequences. However, object segmentation in still images is more difficult. Without motion cues, other cues must be found. Edge detection algorithms are able to extract object contours from images, and were once thought to hold promise for object segmentation in still images. However, additional processing is needed to distinguish between object contours and other edges, such as those produced by textures. The alternative method of region growing (based on luminance or color) has also proven rather ineffective for object segmentation in natural images. In contrast, humans are very successful at object segmentation in still images, suggesting that a model of the early human visual system (HVS) might reveal useful methods for more robust object segmentation in still images. The research results presented in this paper are derived from an HVS model that includes models of Type 1 and Type 2 color contrast cells, and double opponent color contrast cells. By combining the outputs of these cells with edge detected images, object contours can be better distinguished from other contours (such as texture contours and shadow contours) thus providing enhanced object segmentation in cluttered images.
Original language | English (US) |
---|---|
Pages (from-to) | 254-265 |
Number of pages | 12 |
Journal | Proceedings of SPIE - The International Society for Optical Engineering |
Volume | 4662 |
DOIs | |
State | Published - 2002 |
Event | Human Vision and Electronic Imaging VII - San Jose, CA, United States Duration: Jan 21 2002 → Jan 24 2002 |
Keywords
- Color vision
- Computational vision
- Content based retrieval
- Human visual system
- MPEG-4
- Object segmentation
- Primary visual cortex
ASJC Scopus subject areas
- Electronic, Optical and Magnetic Materials
- Condensed Matter Physics
- Computer Science Applications
- Applied Mathematics
- Electrical and Electronic Engineering