This paper presents a person re-identification approach that combines descriptive and discriminative models. The method first uses a feature-based similarity measure to rank samples, where appearance is modeled by region covariance descriptors. Then, a discriminative model is learned using boosting for feature selection, which provides a more specific classifier. The approach is tested on two datasets, showing that combining a generic descriptive model with a discriminative model significantly improves performance compared to individual models. The method also compares with state-of-the-art approaches on a public benchmark dataset. The system consists of a descriptive person model and a discriminative person model. The descriptive model uses a region covariance descriptor to capture visual appearance information, while the discriminative model uses boosting for feature selection to learn a more specific classifier. The descriptive model is based on hand-designed features, while the discriminative model is learned from training data. The system is evaluated on two datasets: the VIPeR dataset (single-shot scenario) and a multi-shot dataset. The results show that combining both models leads to better performance than using either model alone. The method is effective in handling varying appearances, occlusions, and similar instances. The experimental results demonstrate that the proposed approach achieves competitive results compared to state-of-the-art methods. The system is particularly effective in capturing intensity changes between upper and lower body parts and color information, which are crucial for person re-identification. The approach is robust to viewpoint changes and illumination variations, and the use of covariance features provides a compact and effective representation of visual information. The method is efficient and practical for real-world surveillance applications.This paper presents a person re-identification approach that combines descriptive and discriminative models. The method first uses a feature-based similarity measure to rank samples, where appearance is modeled by region covariance descriptors. Then, a discriminative model is learned using boosting for feature selection, which provides a more specific classifier. The approach is tested on two datasets, showing that combining a generic descriptive model with a discriminative model significantly improves performance compared to individual models. The method also compares with state-of-the-art approaches on a public benchmark dataset. The system consists of a descriptive person model and a discriminative person model. The descriptive model uses a region covariance descriptor to capture visual appearance information, while the discriminative model uses boosting for feature selection to learn a more specific classifier. The descriptive model is based on hand-designed features, while the discriminative model is learned from training data. The system is evaluated on two datasets: the VIPeR dataset (single-shot scenario) and a multi-shot dataset. The results show that combining both models leads to better performance than using either model alone. The method is effective in handling varying appearances, occlusions, and similar instances. The experimental results demonstrate that the proposed approach achieves competitive results compared to state-of-the-art methods. The system is particularly effective in capturing intensity changes between upper and lower body parts and color information, which are crucial for person re-identification. The approach is robust to viewpoint changes and illumination variations, and the use of covariance features provides a compact and effective representation of visual information. The method is efficient and practical for real-world surveillance applications.