利用改进的最优聚类算法边缘提取方法研究

ON EDGE DETECTION METHOD USING IMPROVED BEST CLUSTERING ALGORITHM

研究灰度图像的边缘提取的问题。针对传统边缘提取方法容易受到噪声干扰的问题,提出一种利用像素局部方差、信息熵、梯度和分散度特征的聚类算法,并利用Silhouette准则自动测定最优的聚类个数,从而有效地提高聚类和边缘提取的准确性。首先,利用对图像进行预处理,通过对各个像素提取四种不同的特征值,作为聚类分类器的输入;然后,遍历不同的聚类个数,并以Sil-houette作为最优聚类个数的判别标准,最终确定K聚类算法的类别个数。该方法可以有效地提取图像的边缘,尤其对噪声较多的图像能保证很好的边缘提取准确率。

Edge detection issue of greyscale image is studied in the paper. Aiming at the problem of traditional edge detection method that it is prone to noise interference, we propose a clustering algorithm utilising local variance of pixels, information entropies, gradients and dispersion characteristics, and use Silhouette criterion to automatically measure the best clustering number, therefore effectively improve the ac- curacy of clustering and edge detection. First, we pre-process the image and extract four different feature ~~alues on every pixel as the input of clustering classifier. Secondly, different clustering numbers are traversed, and we use Silhouette as the judging criterion of best clustering number, and at last we determine the category number of K-means clustering algorithm. Our method can effectively detect the edge of image, in particular the image with more noises, and can ensure the fine accuracy rate of edge detection.