一种曲面拟合图像边缘特征提取算法

Novel curve fitting edge feature extraction algorithm

针对传统最小二乘支持向量回归函数曲面拟合边缘特征提取算法推广性差的问题,提出首先在模糊特征平面对图像进行模糊去噪增强,使图像各种边缘信息凸显,弱化非边缘信息;然后对图像进行最小二乘支持向量回归函数曲面拟合,对拟合函数求导确定边缘.在对每个点采用相同惩罚因子时,保证了图像中每一像素的邻域最佳函数拟合.仿真实验表明,该算法边缘提取质量清晰细致,效果较好,且各种参数的选择不需人为调节,适合图像预处理阶段应用.

The edge contains much visual information of the image, so the image feature extraction is important in image processing. In this paper, the former least squares support vector machines（LS-SVM） edge feature extraction algorithm is analysed, and it is found that its universality is weaken. So this paper proposes a novel method for edge extraction, in which firstly the digital image is transfered to the fuzzy characteristic plane, where the image edge part is extruded, and the other part is weakened. The the image intensity surface is well fitted by the LS-SVM function, in which the first and second derivatives are calculated. Finally, the rather fine image edge feature can be gained. Experiments show that this algorithm can lead to a higher segmentation quality and that the parameters can be fixed, which is very useful in image processing.