A distributed correlation method for facial expressions and features recognition.

摘要

Given a template image, such as a face image, whether the template object exists in the image database can be decided by using a correlation-based method. Extremely time consuming operations, however, are involved in searching and identifying the image template, especially in a centralized computing environment. With the advantages of distributed computing, such as computational time sharing and data storage sharing, the limitations of image retrieval from a centralized image database can be eliminated. In this thesis, an efficient and computationally fast method is presented to determine whether a patch of face or a face image is stored in a collection of known faces. If such an image exists, a match report is presented. Furthermore, we show that matched image template position and direction of rotation are also determined by the proposed method. We also show how combining the proposed correlation-based image method with distributed computing greatly enhances the performance of face image template searching and recognition. We show the robustness of the proposed method to off-the-plane face image template rotation, image side movements, image resolutions, image occlusions, and varying facial expressions. One of the longer range goals of this work is face analysis and understanding without the participant's cooperation and knowledge. Experimental results are presented to show the potential of the proposed method.