A framework for image enhancement via contextual information and epitome-based representation

• Main Authors: Chee, Seng Chan, Liu, Honghai, Tsz Ming, James Hui, Brown, David J., Khalid, Marzuki
• Format: Conference or Workshop Item
• Language: English
• Published: 2005
• Subjects: TK Electrical engineering. Electronics Nuclear engineering
• Online Access: http://eprints.utm.my/id/eprint/1856/1/marzuki05_framework_image_enhancement.pdf; http://eprints.utm.my/id/eprint/1856/

Image enhancement in our understanding includes quality improvements and understanding improvement of a digital image or video. Our study focuses on the latter; this paper proposes a framework for image understanding using epitomes and contextual information in human-motion involved object recognition systems. The proposed has been raised when considering image enhancement with limited computational resources. We demonstrate the proposed with an example of recognizing a target object, i.e., a watch, on a walking man. First, we employ unsupervised learning to train the video in terms of epitomes, and then the classification of the watch is formulated as a search problem of finding the target pixels in the epitomes. Though the quality of the regenerated object is relatively low, the watch can be recognized with the aid of recent results from human motion analysis.