Autonomous Visual Perception with Cellular Neural Networks

• Main Authors: An-te Liu, 劉安特
• Other Authors: Wei-song Lin
• Format: Others
• Language: zh-TW
• Published: 2004
• Online Access: http://ndltd.ncl.edu.tw/handle/33948876557282968903

碩士 === 國立臺灣大學 === 電機工程學研究所 === 92 === Autonomous visual perception is an essential ability of the early biological visual system. Visual attention, which is a static feature process, and visual alerting, which is a dynamic feature process, constitute the main functions of autonomous visual perception. They enable biological vision to detect rapidly the interested parts in a given scene and automatically select the attentive spots. This mechanism provides an efficient procedure to screen and reduce the image data before commencing further complex image processing and computation. The selected attentive spots usually correlate with the conspicuous parts of the scene, which may be the potential targets. Inspired by the physiological phenomena of the early visual system and current technological visual perception, this research has developed a computational model to extract static and dynamic features for performing autonomous visual perception. Particularly, intention that is specified by the interested optical intensity, color index, or center location has been considered in the extraction to determine the priority of the selected attentive spots. The huge computational burden in the Gaussian and Gabor filtering of the autonomous visual perception system is relaxed by employing the cellular neural networks. The feasibility of the cellular neural network based feature extraction has been verified by computer simulation. Three experiments examining the effectiveness of the design on performing visual attention, visual alerting and visual perception have been conducted and the results are shown.