基于特征匹配要求的纹理边缘检测

为了更精确地进行机器人双目视觉中的特征匹配,需要获得数字图像中丰富而精确的纹理与边缘信息。为此提出了一种新的基于稳态小波变换（SWT）的多尺度纹理边缘检测算法。利用稳态小波变换（SWT）的平移不变性来精确捕获奇异点,在多尺度上通过小波参数的梯度变化检测纹理边缘,而后进行多尺度融合并运用形态学方法得到单像素宽度、精确的纹理边缘二值图。通过对比实验结果表明,相对于经典的边缘检测算法,该算法在抗噪声能力,物体表面纹理的丰富度以及边缘定位的准确性上有很大的提升,并满足特征匹配的需求。

Texture Edge Detection Based on Requirements of Feature Malching

To gain more accurate feature matching of robot binocular vision,accurate and rich texture and edge information in the digital images is needed.Thus,this paper brings a new multi-scale texture edge detection algorithm based on the stationary wavelet transform （SWT）.This algorithm can capture singular point with the shift-invariance of SWT and detect texture and edge via the gradient changing of multi-scale wavelet coefficients,and then fuse multi-scale texture and edge and use the morphological approach to get single-pixel width,precise texture and edge binary map.By comparing the experimental results show that compared with the classical edge detection methods,this new algorithm obtains a great improvement in the anti-noise capability,the richness of object surface texture and the positioning accuracy of edge.