基于生物视觉特征的目标轮廓提取算法

自然场景中的目标轮廓提取是计算机视觉中的一个重要研究问题。其难点在于场景中大量的纹理边缘严重地干扰了轮廓提取的完整性。近年来,一些研究工作将生物视觉特征引入图像边缘轮廓提取,取得了一定的效果。其中通过引入视觉外区抑制特征可以在提取物体轮廓边缘的同时抑制一定量的纹理边缘,从而得到轮廓边缘集合。然而在整合轮廓边缘时,传统模型仅仅采用求交并集的简单合并方法,使得强响应的细小纹理残留。基于此,提出了一种改进的基于生物视觉特征的自然场景目标轮廓提取算法。首先采用多水平抑制方法得到候选轮廓边缘集合。接着将一种基于生物视觉特征的边缘组合方法用于将候选边缘整合成为一个完整的目标轮廓。与传统的外区抑制算法相比,基于视觉特征的轮廓提取算法提高了自然场景中目标轮廓提取的准确性和完整性。

Object Contour Extraction Algorithm Based on Biological Visual Feature

Object contour extraction from natural scenes plays an important role in computer vision.However,it is difficult to preserve the integrality of the object contour in cluttered scenes because of non-meaningful edges engendered from texture field.Recently,the task benefits from a biologically motivated mechanism called as surround suppression (SS) that can preserve the object boundaries while suppressing the texture edges.Nevertheless,the traditional models just adopt a simple combination method of intersection and union that fails to process the short edges with intensity response.This paper proposed an improved natural images object contour extraction algorithm based on biological visual feature.Firstly,a candidate edge set is obtained by multi-level suppression method.Secondly,an edge combination methodbased on biological visual feature is used to combine candidate edges to a completed contour.Experiments show that the proposed method improves the accuracy and integrality compared to the traditional surround suppression methods.