Abstract
In this paper, a new method for shape-based image retrieval is presented. We first identify the dominant points of an object. Then, the geometric moment and perimeter (relative to a referenced dominant point) for each dominant point are computed. A synthetic spectrum plotted from the normalized geometric moments vs. normalized distances is obtained. The area covered by and the perimeter of the spectrum are computed. These two values are used as similarity measures for fast indexing in a shape-based image database. The fine-grained comparisons, based on cross-sectional area and perimeter distribution, are performed on the candidates to select the best matching category. Our method can satisfy necessary requirements of cognitively similarity measures from visual perception, such as rotation, scaling and shearing invariance.
Original language | English (US) |
---|---|
Pages (from-to) | 39-44 |
Number of pages | 6 |
Journal | IEE Conference Publication |
Issue number | 2005-10882 |
State | Published - 2005 |
Event | IEE International Conference on Visual Information Engineering, VIE 2005 - Glasgow, Scotland, United Kingdom Duration: Apr 4 2005 → Apr 6 2005 |
All Science Journal Classification (ASJC) codes
- Electrical and Electronic Engineering
Keywords
- Chain Codes
- Dominant Points
- Shape Retrieval
- Similarity Measures
- Spatial Feature