Viewpoint invariant pedestrian recognition is a challenging problem in computer vision due to the difficulty of matching objects with unknown viewpoints and poses. This paper presents a method using an ensemble of localized features (ELF) to achieve viewpoint invariant recognition. Instead of designing specific features manually, the method defines a feature space based on problem intuition and uses machine learning to find the best representation. AdaBoost is used to learn both object-specific representations and discriminative recognition models, allowing simple features to be combined into a single similarity function. The method is evaluated on a viewpoint invariant pedestrian recognition dataset, showing superior performance compared to previous benchmarks in both recognition and reacquisition. Pedestrian tracking is challenging, especially with increasing numbers of targets and occlusions. However, the ultimate goal of surveillance systems is to understand scenes and provide effective operator interfaces, which requires searching camera networks for people of interest. This is effectively pedestrian re-identification without temporal constraints. Pedestrian recognition faces challenges beyond tracking, particularly the lack of temporal information, requiring matching decisions based solely on appearance models. Traditional methods like templates and histograms have limitations, especially with illumination changes. This paper proposes a hybrid model combining templates and histograms, constructed using machine learning to maximize discriminability for training data. The learned model is an ensemble of localized features, providing a similarity function for comparing pedestrian images, useful for both re-identification and recognition. The method is evaluated on the VIPeR dataset, showing superior performance.Viewpoint invariant pedestrian recognition is a challenging problem in computer vision due to the difficulty of matching objects with unknown viewpoints and poses. This paper presents a method using an ensemble of localized features (ELF) to achieve viewpoint invariant recognition. Instead of designing specific features manually, the method defines a feature space based on problem intuition and uses machine learning to find the best representation. AdaBoost is used to learn both object-specific representations and discriminative recognition models, allowing simple features to be combined into a single similarity function. The method is evaluated on a viewpoint invariant pedestrian recognition dataset, showing superior performance compared to previous benchmarks in both recognition and reacquisition. Pedestrian tracking is challenging, especially with increasing numbers of targets and occlusions. However, the ultimate goal of surveillance systems is to understand scenes and provide effective operator interfaces, which requires searching camera networks for people of interest. This is effectively pedestrian re-identification without temporal constraints. Pedestrian recognition faces challenges beyond tracking, particularly the lack of temporal information, requiring matching decisions based solely on appearance models. Traditional methods like templates and histograms have limitations, especially with illumination changes. This paper proposes a hybrid model combining templates and histograms, constructed using machine learning to maximize discriminability for training data. The learned model is an ensemble of localized features, providing a similarity function for comparing pedestrian images, useful for both re-identification and recognition. The method is evaluated on the VIPeR dataset, showing superior performance.