Back to results list
Show full item record
Please use this identifier to cite or link to this item:
|Title:||iJade face recognizer : a multi-agent based pose and scale invariant human face recognition system||Authors:||Ao Ieong, Wai-heng Tony||Degree:||M.Phil.||Issue Date:||2006||Abstract:||Human face recognition is a challenging pattern recognition problem. Problems with current face recognition systems include the difficulty of recognizing faces presented at various angles, at various scales, with varying facial expressions, under uneven illumination, and in cluttered scenes. This research proposes an automated face recognition system that attempts to deal with these difficulties at each stage of the face recognition processes: face detection, feature extraction, and face identification. In face detection, we deal with the issue of detecting and localizing faces in a cluttered scene. We begin by identifying possible skin regions using a Gaussian mixture model of human skin tones on the HSV color space. We then use Neural Networks to classify and detect faces within these skin regions at a variety of image scales and use a committee network to detect face with different poses. In the next step, feature extraction, we use Gabor features derived from Gabor wavelet representation to extract the most salient, distinctive, and invariant facial features. We adopt the Gabor feature extraction technique because these features are robust to change in illumination and facial expression. To identify faces presented at a variety of angles, we adopt template-based Gabor Eigenface features and feature-based Gabor Labeled Elastic Graph face features. The third and final step, face identification attempts to determine the identity of a query face by comparing it with a face database. At this stage, the search operation is conducted using agent technology and agent technology allows the system to operate at various scales and on various face models without any deterioration in the system performance. The proposed approach was tested on a variety of images and face databases and compared with other face recognition models to show the benefits and improvement. This research was implemented on iJADE Face Recognizer, a multi-agent based pose and scale invariant human face recognition system. iJADE (intelligent Java Agent Development Environment) is an intelligent agent-based platform for supporting the implementation of artificial intelligence (AI) functionalities. Each of the three major processes in the proposed face recognition system has been designed in a separate module and each module is implemented using intelligent agents. Agent technology provides a collaborative platform that can be used to build sophisticated applications and allows different agents to interact with each other. The performance of the iJADE Face Recognizer is superior to that of other approaches because the agent platform operates in an asynchronous runtime environment which enables parallel processing and under a distributed architecture supports a number of varied operating scales. iJADE Face Recognizer has numerous potential applications. For example, as an automatic human face surveillance system, it can automatically analyze scenes and extract human faces in complex environments and can efficiently and effectively perform invariant human face feature extraction, identification and recognition.||Subjects:||Hong Kong Polytechnic University -- Dissertations.
Human face recognition (Computer science)
|Pages:||xiv, 96 leaves : ill. (some col.) ; 30 cm.|
|Appears in Collections:||Thesis|
View full-text via https://theses.lib.polyu.edu.hk/handle/200/2512
Citations as of Nov 22, 2023
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.