噪声环境下基于不确定性度量的随机HT直线检测

UNCERTAINTY MEASUREMENT-BASED RANDOM HOUGH TRANSFORM LINE DETECTION IN NOISY CONDITION

含噪图像中直线的自动检测是机器视觉和图像处理中的热点问题之一。基于Hough变换的直线检测算法中采用硬投票方案,在噪声环境下检测精度下降且占用内存大。为了提高检测算法的抗噪性和降低算法的计算复杂性,提出了一种新的将边缘点不确定性度量和随机Hough变换相结合的直线检测算法。该算法在所建立的点属于某直线上不确定性度量概率模型基础上,根据随机选择的两点间直线参数,按照Bayesian法则用基于不确定度量的参数空间软投票提高了检测算法的抗噪能力。实验结果表明,算法在较高的噪声(方差大于0.03)时,检测误差小于1‰,检测时间是单纯不确定度量直线检测方法的1/2,比传统Hough变换算法快10-15倍。

Automatic line detection in noisy images is one of the research hot-spots in computer vision and image processing. Conventional Hough Transform （HT） algorithm uses brute-force voting scheme,its detection precision will degrade especially in noisy images, and occupies large amount of memory. In order to improve anti-noise performance of the detection algorithm and to reduce the computational complexity, a new line detection method is proposed, which integrates uncertainty measures of edge points and randomized Hough transform. After building a probability model of measurement of the uncertainty that each edge point belongs to a line, the algorithm computes line parameter between two points randomly selected in edge images. An uncertainty measurement-based＂ soft voting scheme＂ in parameter space is utilized according to Bayesian rules to improve anti-noising ability of the detection algorithm. Experimental results show that the proposed algorithm has the detection error rate less than 1‰ under relative higher noise level （ noise variant larger than 0.03 ） but its detection time is only 1/2 of the pure uncertainty measurement line detection method used ,and is 10-15 times faster than conventional HT method.