摘要
本文提出了一种多方向通道的立体匹配算法。该算法模拟人类视觉机制,将视觉的各向异性特征、多方向通道特性引入体视匹配过程,首先将原始像对分解成不同的方向通道,进而在各个方向通道中并行地进行基元提取与立体匹配。由于在每个方向通道中仅包含了某一特定方向区间内的图像信息,在方向通道中进行特征提取与匹配具有定位精度高、抗噪能力强、匹配性能稳定等优点。实验结果表明,采用本算法对实际场景进行体视匹配具有较好的准确性和可靠性,且适于并行处理。
In this paper, a stereo matching algorithm for multi-directional channels is proposed, which imitates the human visual mechanism, and brings the anisotropic information and multi-directional channel character of vision into the area of stereo matching. First, the original stereo images are decomposed into several directional channels, and then the primitives extraction and matching are performed parallelly in each pair of channels. Since the information in each channel belongs only to a limited interval of direction, the feature extraction and matching in each pair of directional channels are rather precise, noiseproof and stable. Experimental results show that the proposed algorithm is precise and robust in coping with real scenes, and suited for implemented parallelly.
出处
《模式识别与人工智能》
EI
CSCD
北大核心
1994年第3期208-214,共7页
Pattern Recognition and Artificial Intelligence
基金
国家教委博士点资助项目
关键词
机器视觉
立体匹配
算法
多方向通道
Matching Algorithm, Computer Vision, Directional Channel.
