摘要： 将已有改进的Harris 角点检测方法应用于图像拼接时,在单一尺度下检测角点会存在角点信息丢失、角点位置不准确和对噪声敏感致使检测率不高等缺点。为此,结合双边滤波和多尺度检测,提出一种Harris 角点检测改进方法。采用双边滤波器代替原有的高斯低通滤波器,以增强方法的鲁棒性,引入近邻传播聚类方法将图像分块,避免因阈值问题造成角点分布不均,并将多尺度概念引入到改进方法中,判断候选角点是否为真实角点,剔除伪角点,使得角点检测更加精确。在VS2010 + OpenCV 平台上进行实验,结果表明,与基于图像分块的多尺度Harris 角点检测方法相比,该方法在去伪角点、漏检、准确度等方面取得了更好的效果,且具有更好的抗噪性。
关键词:
Harris 方法,
双边滤波,
多尺度,
图像分块,
角点检测
Abstract: When applying in the image stitching,some improved Harris corner detection methods still have defaults like
only detecting corners in a single scale,the lost of corner detection information,unaccurate angular position and low detection rate caused by the sensitivity to noise. Combining bilateral filtering and multi-scale testing,this paper proposes an improved Harris corner detection method. The method uses a bilateral filter instead of the original Gaussian low-pass filter to enhance the robustness. The introduction of affinity propagation clustering method to block image avoids the uneven distribution of the corner caused by the threshold problems. The introduction of multi-scale concept to the improved method determines whether a candidate corner is true corner and excludes false corners,and makes corner detection more precise. Experiments are achieved in the platform of VS2010 + OpenCV. Experimental results show that, compared with the effect of the current corner detection method,the improved Harris corner detection method proposed in
this paper can detect the image corners more accurately,achieves better effect and is more practical.
Key words:
Harris method,
bilateral filtering,
multi-scale,
image block,
corner detection
中图分类号: