图像3D轮廓提取

Extracting 3D Contours From Image

针对现有3D轮廓提取方法计算量大、图像立体匹配难、图片中含有大量噪音的问题,提出了一种利用早期认知视觉系统提取图像3D轮廓的方法。早期认知视觉是连接早期视觉和认知视觉的中间平台,通过早期认知视觉系统提取出图像的2D基元,2D基元是图片小块,有丰富的语义信息包括位置,方向,相位和三色值;再由两个立体图像的相一致的2D基元配对来构建3D基元,通过3D基元的共面性和共色性得到图像的3D轮廓。在Ubuntu10.04下,通过输入立体图片对,对这种方法进行了实验,实验结果表明图像的3D轮廓能完好地提取出来,有效地保留图像的必要信息并具有强抗噪性。

The existing approaches to extract 3 D contours have drawbacks such as huge computation,difficulty in matching the stereo images and the large amount of noise in images.To solve these problems,this paper presents a method that use the early cognitive vision system to extracting 3 D contours of an image.The Early Cognitive Vision System(ECV)is an intermediate stage that mediates information between the Early Vision stage and the Cognitive Vision stage.2 D primitives are extracted by ECV from images.A 2 D primitive represents a small image patch in terms of position,orientation,phase and three color values;pairs of corresponding 2 D primitives across two stereo views afford the reconstruction of a 3 D primitive.According to the co-planarity and co-colority between 3 D primitives,the 3 D contours of the image can be obtained.In Ubuntu10.04,the method by inputting two stereo image pair is tested,the experiment result shows the 3 D contours of image completely,that reflect the necessary information of the image and the method is insensitive to noise.