Efficient schemes for indexing and retrieval from large face databases

Abstract

The aim of this research is to develop efficient techniques for face recognition with a large face database. In practice, the number of faces in a database may range from hundreds to several tens of thousands. As a result, many problems need to be considered when developing a practical human face recognition system. One of these problems is the required search time for a human face in the large database. To reduce the search time, an efficient indexing and retrieval algorithm is required. In this thesis, efficient and accurate face recognition techniques based on a 3-D face structure and the optimal selection of Gabor features will be investigated for a large human face database. In this thesis, a new algorithm is proposed to derive the 3-D structure of a human face from a group of face images under different poses. Based on the corresponding 2-D feature points of the respective images, their respective poses and the depths of the feature points can be estimated based on measurements using the similarity transform. To accurately estimate the pose of and the 3-D information about a human face, the genetic algorithm (GA) is applied. Our algorithm does not require any prior knowledge of camera calibration, and has no limitation on the possible poses or the scale of the face images. It also provides a means to evaluate the accuracy of the constructed 3-D face model based on the similarity transform of the 2-D feature point sets. We have shown that our approach can be applied to face recognition such that the effect of pose variations can be alleviated. Experimental results show that our proposed algorithm can construct a 3-D face structure reliably and efficiently. Another approach to enhance the performance of face recognition in a large face database is the use of selective Gabor features. Gabor features have a good performance level for face recognition. However, the extraction of Gabor features at different centre frequencies and orientations is computationally intensive. An algorithm to extract and select the Gabor features for face recognition has been proposed. The Gabor features of the images are extracted using a simplified version of the Gabor wavelet; this can reduce the extraction runtime by 30% compared to using original Gabor wavelets. As the responses of the Gabor wavelets are strongly related to those edges that are perpendicular to the wave vectors, edge detectors with different orientations are employed. To further reduce the recognition runtime, the size of the database can be decreased so that the number of comparisons between the query image and each model image can be reduced. Experimental results show that the recognition rate can be maintained with a faster processing time.