为了提高角点检测算法的精确度,同时保持较低的时间复杂度,提出了一种基于自定义的局部角点响应显著度和迭代分割逼近的角点检测方法。首先,定义了一个新的测度量——局部角点响应显著度(LCRS),用来衡量一个候选角点在其局部区域内的响应显著程度,并证明了基于LCRS的角点检测准则等价于局部自适应阈值法。其次,将LCRS视作区域的响应显著度性质,把寻找角点的任务转化为寻找高显著性区域。据此,提出迭代分割的策略用来逐步收缩显著区域,最终逼近真实角点的位置。迭代分割逼近(ISA)算法可以使用不同的角点响应函数(CRF)定义,而且其平均情况时间复杂度与Harris算法相同。实验结果表明,当采用Noble算子的CRF时,ISA算法平均误检率、漏检率分别比Noble算子低4.62%和5.59%;而当采用Harris算子的CRF时,这两个比率也分别比Harris算子低2.87%和3.37%。而且这两种情况下ISA算法的平均运行时间均小于Harris算子和Noble算子。

To improve detection accuracy and speed,a corner detection method based on the novel Local Corner Response Saliency (LCRS) and Iterative Segment Approaching (ISA) is proposed.First,a new metric,the Local Corner Response Saliency,is defined to measure the response saliency of a candidate corner point within the local region where it is located.The corner detection criterion based on LCRS proved to be equivalent to a local adaptive thresholding method.Then,by treating LCRS as a property of response saliency of a region,the task of searching corner points is transformed to the task of searching special regions with high saliency.With this point of view,an iterative segment strategy is suggested to gradually narrow down these salient regions and finally approaching the actual positions of the true corners.The ISA algorithm can work with various definition of corner response function (CRF),and its speed in average case is identical to the Harris detector.Experiments show that when using Noble’s CRF,the average false and miss detection rates of ISA are lower than those of Noble detector by 4.62% and 5.59%,respectively.When using the Harris’ CRF,these tow rates are also lower than those of the Harris detector by 2.87% and 3.37%,respectively.Moreover,in both cases the average running time of ISA is shorter than these of the Harris and Noble detectors.