视觉感知结合学习的自然图像轮廓检测

本文结合非经典感受野的视觉特性与机器学习的方法,提出了一种自然图像轮廓检测模型.当非经典感受野中的刺激与感受野中心刺激形成一种精确的空间结构时,将对中心产生一种增强效应;另一方面非经典感受野中抑制作用会降低同质成分的响应,我们将这两个机制分别用于增强光滑的轮廓和减少背景中与结构无关的干扰成分.利用逻辑回归概率模型将感受野中的信息与来自非经典感受野中的信息进行有效融合,并根据图像的手工标注数据库,通过学习方法获得一组最优的模型参数.自然图像的实验结果表明该轮廓检测方法能极大地抑制来自纹理的局部边缘,减少虚假轮廓,同时能增强具有一致空间结构的成分,避免轮廓缺失.最后利用Berkeley图像数据库定量地评价了我们方法的性能,并与相关方法进行了比较.该模型不仅为复杂场景中的轮廓检测提供了一个可行的策略,并有助于对生理视觉机制的理解.

Detecting natural image contours by combining visual perception and machine learning

A model for detecting contours in natural images is presented by combining the visual perceptual mechanisms of nonclassical receptive fields and machine learning. When the stimuli inside and outside the receptive field （RF） form a precise spatial configuration, the surround stimuli will enhance the response of the central stimulus~ on the other hand, surround inhibition will reduce the responses to homogeneous elements. The two different visual mechanisms are used to enhance the well-organized structures and reduce the non-meaningful distractors engendering from texture fields, respectively. We approach the task of information combination as a supervised learning problem using the logistic regression model, where we will learn the combination rules from the ground truth data. Our experiments demonstrate that the model can dramatically suppress texture edges and reduce spurious contours, and meanwhile can enhance the responses to the elements with coherent spatial configurations and avoid ground-truth contours missed by the detector. Finally, we evaluate quantitatively our results with the Berkeley segmentation dataset and compare with related work. The proposed method not only provides a biologically feasible scheme for contour detection but also contributes to further understand visual mechanisms.