限制型自适应SUSAN边缘检测算法

对于边缘检测中传统SUSAN(smallest univalue segment assimilating nucleus)算法,固定门限会将非边缘点划入核值相似区(univalue segment assimilating nucleus, USAN),并经过单一阈值判断,非边缘点易被误判为边缘点,导致算法的低鲁棒性.针对此问题,提出了结合自适应门限算法和阈值选择策略的限制型自适应SUSAN算法.首先,分析SUSAN算法优缺点,根据USAN特点以及同异侧噪声容忍度范围设置阈值选择策略,减少误判并提高噪声鲁棒性;然后采用与USAN内像素值正相关的自适应门限算法,进一步增强边缘检测能力.在标准测试图像以及不同类型噪声的经典灰度图中实验结果表明,相比于传统SUSAN算法和Canny, Prewitt, Sobel, LoG, Roberts等边缘检测算法,该算法在客观图像评价指标FSIM值,PFOM值和准确率上均高于其他算法;而在主观视觉上,在无噪条件下能够更好地抑制纹理区域像素干扰,检测边缘更完整丰富.特别是在大量噪声干扰导致其他算法均失效的情况下,该算法在抑制噪声的同时,仍能有效地检测出图像边缘.

Constraint Self-Adaptive SUSAN Algorithm for Edge Detection

For the traditional smallest univalue segment assimilating nucleus(SUSAN)algorithm in edge detection,the fixed critical value will consider the non-edge points to be within the univalue segment assimilating nucleus(USAN)and judge by the threshold,resulting in the misjudgment of the non-edge points as edge points and low noise robustness.To this end,we propose a constraint self-adaptive SUSAN algorithm combining the adaptive critical value algorithm and threshold selection strategy.Firstly,the threshold selection strategy is set according to the characteristics of the USAN and the noise tolerance to reduce misjudgment and improve the noise robustness.Then an adaptive critical value algorithm with a positive correlation with the pixel value in USAN is used to enhance the edge detection capability.Compared with the traditional SUSAN algorithm and Canny,Prewitt,Sobel,LoG,Roberts algorithms,the experimental results show that the proposed method has higher precision,FSIM,and PFOM in the qualitative measurement.In the case that lots of noise interferences cause the failure of other algorithms,the proposed algorithm can effectively detect the image edge while suppressing the noise.