Graph embedding techniques in face verification
Face verification system comprises three operational components: preprocessing module, feature extraction/ dimensionality reduction module and verification module. In this study, the dimensionality reduction process is focused. The acquired facial data is usually represented in a high dimensional ve...
Saved in:
| 主要作者: | |
|---|---|
| 格式: | Thesis |
| 出版: |
2012
|
| 主題: | |
| 標簽: |
添加標簽
沒有標簽, 成為第一個標記此記錄!
|
| 總結: | Face verification system comprises three operational components: preprocessing module, feature extraction/ dimensionality reduction module and verification module. In this study, the dimensionality reduction process is focused. The acquired facial data is usually represented in a high dimensional vector carrying highly redundant data. In fact, its discriminative features are embedded on a much lower dimensional manifold. Hence, dimensionality reduction is the crucial module to extract representative features from the face data. There are two important considerations in designing a dimensionality reduction technique in face verification: (1) how to effectively exploit the limited available training samples?And (2) how to seek the most discriminative facial feature representations? Graph embedding approach, which is also known as linearization of graph
embedding, is a relatively new emerging technique in dimensionality reduction process. This approach seeks underlying data structures by modelling local manifold structure based on data similarities on an affinity graph via manifold preserving criterion. In this thesis, three graph embedding techniques are proposed for face verification. |
|---|