基于信息测度和核函数极限学习机的图像边缘检测

IMAGE EDGE DETECTION BASED ON INFORMATION MEASURE AND KERNEL LEARNING MACHINE

图像的边缘检测技术是机器视觉中图像识别、图像分割与处理以及模板匹配的基础。针对传统边缘检测算子的检测精度有限,对噪声的敏感度较高的问题,提出一种基于信息测度和核函数极限学习机(KELM)的图像边缘检测方法。该方法构造一个描述边缘点信息测度的特征矢量,将特征矢量样本数据集对核函数极限学习机(KELM)进行分类训练,实现边缘检测。同时采用度量F评价模型对不同边缘检测方法的性能进行评价。实验结果表明,ISKELM图像边缘检测的效果优于Canny算子、Sobel算子以及ELM图像边缘检测,提取的图像边缘更加清晰,对于噪声的抑制能力更强,虚假边缘大大减少。

Image edge detection is the basis of image recognition,image segmentation and processing in machine vision.In view of the traditional edge detection operator,high sensitivity to noise,this paper put forward a kind of extreme learning machine based on information measure and nuclear function(KELM) image edge detection method.It constructed a characteristic vector describing the edge point information measure.The characteristic vector sample data set of kernel function extreme learning machine(KELM) was classified to train,and realized the function of edge detection we used metric F as evaluation index of image edge detection.Experimental results show that KELM image edge detection is better than Canny operator,Sobel operator and ELM image edge detection.The image edge extracted by this method is clearer,with stronger noise suppression ability and greatly reduced false edges.