多水平外区抑制的轮廓检测

Contour detection based on multilevel inhibition

提取自然图像中的物体轮廓是机器视觉研究的重要问题,主要困难在于自然图像中的纹理性边缘严重干扰了物体轮廓的提取。研究表明视皮层方位选择性神经元的非经典感受野机制使得人类视觉系统在处理自然图像时不仅能够抑制纹理性边缘,而且能够增强物体的轮廓。基于此人们提出多种仿生轮廓检测算法,但算法中被称为抑制水平的参量在取值较高时会漏检部分轮廓,而在其取值较低时又会引入过多的纹理性边缘。针对这一问题,提出多水平外区抑制轮廓检测算法,通过整合各级单水平外区抑制的检测信息,有效抑制了纹理性边缘和降低了漏检轮廓的可能性。实验结果表明,相对于传统算法,新算法在轮廓检测性能上提高了10%左右,并具有更好的稳健性。

Detecting object contours from natural images plays an important role in machine vision. However, because of the texture edges existing in natural images, it becomes very hard to implement. Relevant research on orientation selective neurons in the primary visual cortex shows, that a mechanism, called non-classical receptive field, can inhibit texture edges and facilitate isolated edges when the visual system processes natural images. Many biologically motivated models have been proposed for contour detection, but they share a common problem which is that some contour elements will be lost if the value of inhibition level is set to high, while some texture edges will be retained if it is set to low. In order to solve this problem, we present a new model, which combines the information from different inhibition levels. It effectively suppresses texture edges and reduces the possibility of losing contour elements. Experimental results show that in comparison with the traditional algorithms, the new algorithm increases performance about ten percent and is more robust.