Face recognition enhancement by employing facial component classification and reducing the candidate gallery set

摘要

Face recognition systems constitute a significant proportion of robotic systems. Learning algorithms such as deep learning and machine learning provide state-of-the-art algorithms with highly improved recognition rates. A majority of these algorithms convert a face image into a feature matrix, holistic or local, matched against all the feature matrixes in gallery for recognition. However, there is a lack of algorithms that consider spatial aspects of facial components such as eyes, nose, etc. for face matching. We propose a novel methodology which allows us to reduce the candidate gallery set based on spatial aspects of facial components. We classify faces based on spatial aspects of facial components, reducing the gallery size significantly. The proposed method can be employed independent of the recognition algorithm used and improves recognition rate. Experiments were performed on the CMU-PIE face image database, and PCA was used as a basic face recognition algorithm.