层次聚类算法的实时图像边缘检测及FPGA实现

Hierarchical Clustering Algorithm of Real-Time Image Edge Detection on FPGA

Sobel、Roberts算子是基于微分得出的,由于模板和阈值固定,因此缺乏自适应性。将采集到的实时灰度图像先进行中值滤波,使用分裂聚类算法对实时灰度图像梯度值进行第1次聚类,然后对第1次分裂聚类的结果进行凝聚聚类,再进行第2次的分裂聚类,最后对第2次聚类的结果进行自适应阈值判断得出图像边缘,并在FPGA上实现。实验结果表明,采用层次聚类算法检测出的边缘更加精细,抑制噪声能力更强。

The classical edge operators are mostly based on the gradient computation such as Sobel, Roberts operator and so on, which has features of poor self-adaptability due to its fixed template and threshold. In this paper, the noise in real-time gray-scale image is filtered by median filtering, and then the gradient value of real-time gray-scale image is classified for the first time by split-level clustering algorithm. After that, this gradient is classified for the second time by agglomerative clustering algorithm. The image edge is obtained by using adaptive threshold. The results demonstrate that the edge detected by the hierarchical clustering algorithm is more accurate, and the background noise is filtered as well.