一种局部最佳阈值预测的自适应角点检测方法

摘要

为解决图像角点检测阈值选取方法计算量大的问题,提出一种新的自适应角点检测方法.分析能够反映图像灰度分布、对比度和相关性因素的9个基本统计特征,通过提取4 848幅样本图像的基本统计特征,并按主成分分析方法计算4项反映图像不同属性的综合指标.建立多元非线性局部最佳阈值预测模型,由训练数据对模型参数进行优化估计,得到指导角点检测自适应阈值选取的预测模型.实验结果表明,预测模型的引入能够改善图像显著角点检测质量,与原始检测算法相比,复杂图像中显著角点检出率平均提高45％,非显著角点误检率平均降低81％.%In order to solve the problem of large computational load of the corner detection threshold selection method,a new adaptive corner detection method is proposed.Nine basic statistical characteristics that can reflect the gray distribution,contrast and correlation of the images are analyzed.The basic statistical characteristics of 4 848 samples are extracted,and the principal components analysis is used to calculate 4 comprehensive indexes reflecting the different attributes of the images.The multivariate nonlinear local optimal threshold prediction model is established,and the model parameters are optimized and estimated by the training data.The prediction model of the guidance corner detection adaptive threshold selection is obtained.Experimental results show that the introduction of prediction model can improve the quality of detection of significant corners of the image,detection rate of significant corners in complex images is improved by 45％ on average compared with the original detection algorithm,and the average false detection rate is reduced by 81％ on average.