【摘要】 为了能在模糊或有噪声的图像中直接检测角点,提出了一种对比多尺度下灰度差异几何特征和搜索局部灰度最大权值的角点检测方法。该方法根据图像中角点位置的灰度差异,采用多尺度圆形检测器遍历图像,对比不同尺度下检测器所得到的几何参数来初步确定角点位置并剔除边缘附近的伪角点;根据局部灰度权值最大原则去除由于圆形检测器的离散性而聚集在真角点周围伪角点,并确定角点的最终位置。实验结果表明,相对于传统算法,该方法能够在对噪声或模糊图像不做预处理的情况下检测出角点。更多还原

【Abstract】 To detect the corner points in blurred or noised image directly,a corner point detection method that includes comparison of geometrical features with grey-level difference under multi-scale and searching of local grey-level weighting maximum is presented.According to the grey-level difference of corner point position in image,circular detectors in multi-scale are adopted to scan the whole image.The geometrical features of corner point gotten from the different scale detectors are compared to confirm corner points position and the fake corners near edges are eliminated.For the problem of fake corner points congregate around true corner point because of the discretion of circular detector,the positions of true corner points are extracted according to the local grey-level weighting maximum rule.The experiment results show that the corner points in blurred or noised image can be efficiently detected with this method.更多还原