The projects described below are typical research projects. The actual projects will be chosen on an individual basis for every student, based on his or her strengths and interests.
Given a set of face images, determine (recognize) which one corresponds to a particular face. This is an important application which requires both low and high-level visual information processing. The standard approach to face recognition is model-based. Typically: (i) we identify face features (landmarks), such as the eyes, the nose, the mouth; (ii) associate the features to (geometrical) models; (iii) determine a face model by finding the corresponding feature models; (iv) match the face model to the faces in the set. Target applications include security and banking identification systems.
Given a series of face snapshots, analyze, interpret, and synthesize facial expressions. This is part of a brod area of research which studies visual behavior as reproduced by artificial systems. For an arbitrary face image (frontal view) sequence, we want to infer semantic or physical attributes. The first stage corresponds to face analysis. This describes the process of decomposing facial characteristics, such as face landmarks and their mutual relations into models. The second stage corresponds to face interpretation. This is determined by the application. For example, if we want to process "emotional states" of a person, we have to compare model(s) obtained at the analysis stage, to models describing facial emotion, such as smile (happy), cry (sad), or grin (disgust). The third stage is to synthesize the various emotional states using computer graphics models. Applications include video teleconferencing and systems which combine auditory and visual information.
Segmentation and Tracking in Dynamic ScenesGiven a video sequence of a dynamic scene we want to segment the objects (persons, cars, etc.) relative to the static background - the figure/ground segmentation problem. Once segmented, we want to track specific objects which are pre-determined by the application being implemented. Segmentation is among the most difficult problems in visual information processing. Motion segmentation is especially important for video applications. Automatic object tracking, is fundamental in applications such as monitoring and surveillance.
Temporal Segmentation of Video
Given a video sequence, determine in which frames scene changes occur. Temporal segmentation is a major task in video production and content-based handling. It may require low and high-level visual information processing, depending on how much is known about the structure of the "world" and the problem being addressed. Target applications include assisted non-linear video editing systems, and video library indexing.
More ....
Facial Extraction and Tracking
© 2001 UNR Computer Science Department. All rights reserved.
Program related questions/comments: bebis@ cs.unr.edu
Last Update: