摘要
轮廓检测是实现基于形状的目标识别任务的关键,然而由于图像背景中存在大量的无关干扰成分,因此要从自然场景中自动地检测出目标的轮廓是非常困难的。人类可以很容易地从环境中识别物体,这得益于人类的视觉感知机制。根据此机制,建立了基于图像切线方向感受野的轮廓检测方法。本方法用Gabor函数模拟感受野中心神经元的响应,非经典感受野的建模充分考虑了邻域刺激与相对位置、距离、朝向之间的关系。本方法与传统的基于人类视觉感知机制方法的最大不同在于图像点的切线方向选择对应朝向的中心神经元,物理意义明确。实验结果表明:该方法能更有效地抑制纹理边缘并减少轮廓自身的破坏,具有更好的检测准确性和稳定性。
Contour detection plays an important role in shape-based object recognition tasks. It is an intractable problem of how to detect the contour of object from natural scenes automatically. Thanks to the human visual perception system, human can identify objects easily. According to this system, this paper presents a contour detection method based on receptive field towards the tangential direction. This method simulated the response function of single visual cell by Gabor function. A new non-classi- cal receptive field model was constructed by fully considering the difference in distance, position and direction between neighboring stimuli and the central neuron. The salient difference between this method and those traditional methods based on human visual perception system is that the preferred o- rientation is selected by the direction of tangent and this gives ＂the preferred orientation＂ a physical
出处
《重庆理工大学学报（自然科学）》
CAS
2014年第4期83-88,共6页
Journal of Chongqing University of Technology：Natural Science
关键词
轮廓
人类视觉感知机制
感受野
切线方向
contour
human visual perception system
receptive field
the tangential direction
