[opencv]K-means聚类算法分割图像

1.K-means算法

K-means是一种经典的无监督聚类算法。

算法原理:

(1)从N个数据对象中取K个点作为聚类中心。

(2)计算N个数据对象到各个聚类中心的距离，将N个数据对象划分到距离最近的中心，形成K个聚类。

(3)计算每个聚类的中心，将新的中心代替原来的中心。

(4)检查新老聚类中心的距离，重复(2)-(3)直到距离小于规定阈值。

2.opencv函数

链接：http://www.opencv.org.cn/opencvdoc/2.3.2/html/modules/core/doc/clustering.html?highlight=kmeans#cv.KMeans2

cvKMeans2(const CvArr* samples, int nclusters, CvArr* labels, CvTermCriteria criteria, int attempts=1, CvRNG*rng=0, int flags=0, CvArr* centers=0, double* compactness=0)

kmeans(InputArray samples, int clusterCount, InputOutputArray labels, TermCriteria criteria, int attempts, int flags, OutputArray centers=noArray() )

Parameters:

samples – Floating-point matrix of input samples, one row per sample. clusterCount – Number of clusters to split the set by. labels – Input/output integer array that stores the cluster indices for every sample. criteria – The algorithm termination criteria, that is, the maximum number of iterations and/or the desired accuracy. The accuracy is specified as criteria.epsilon. As soon as each of the cluster centers moves by less than criteria.epsilon on some iteration, the algorithm stops. attempts – Flag to specify the number of times the algorithm is executed using different initial labelings. The algorithm returns the labels that yield the best compactness (see the last function parameter). flags – Flag that can take the following values: KMEANS_RANDOM_CENTERS Select random initial centers in each attempt. KMEANS_PP_CENTERS Use kmeans++ center initialization by Arthur and Vassilvitskii [Arthur2007]. KMEANS_USE_INITIAL_LABELS During the first (and possibly the only) attempt, use the user-supplied labels instead of computing them from the initial centers. For the second and further attempts, use the random or semi-random centers. Use one of KMEANS_*_CENTERS flag to specify the exact method. centers – Output matrix of the cluster centers, one row per each cluster center.

sample

输入样本的浮点矩阵，每个样本一行

clusterCount

给定的聚类数目 

labels

输出整数向量：每个样本对应的类别标识 

criteria

迭代终止条件：迭代最大次数和精度

attempts

运行K-means的次数，取结果最好的聚类为最终聚类，和flags一起使用

flags

KMEANS_RANDOM_CENTERS 每次尝试选取随机初始化中心

KMEANS_PP_CENTERS  使用Arthur and Vassilvitskii算法进行中心初始化

KMEANS_USE_INITIAL_LABELS使用用户自定义初始点

3.实现

处理图像：

将图像的颜色数据进行K-means聚类，这里有黑、白、红三色，所以分为3类进行效果测试。

将图像每个像素三通道值按行存给样本samples。

	for (int i = 0; i < img->width;i++)
	{
		for (int j = 0; j < img->height;j++)
		{
			CvScalar s;
			s.val[0] = (float)cvGet2D(img, j, i).val[0];
			s.val[1] = (float)cvGet2D(img, j, i).val[1];
			s.val[2] = (float)cvGet2D(img, j, i).val[2];
			cvSet2D(samples,k++,0,s);
		}
	}

nCluster定义为3。

进行K-means聚类分析

	cvKMeans2(samples,nCluster,clusters,cvTermCriteria(CV_TERMCRIT_ITER,100,1.0));

保存结果给clusters,这里是根据图像坐标对应颜色的聚类结果分别标志为0,1,2。

测试分类效果：

将图像基于聚类标签结果进行提取。

(1)提取标签0

	k = 0;
	int val = 0;

	for (int i = 0; i < img->width;i++)
	{
		for (int j = 0; j < img->height;j++)
		{
			val = (int)clusters->data.i[k++];

			CvScalar s1;
			if (val == 0)
			{
				s1.val[0] = 255;
			}
			else
			{
				s1.val[0] = 0;
			}
			cvSet2D(bin,j,i,s1);
		}
	}

将图像基于bin进行mask复制，显示标签0对应的区域

	cvThreshold(bin,bin,200,255,CV_THRESH_BINARY);
	IplImage *img1=cvCreateImage(cvSize(img->width,img->height),img->depth,img->nChannels);
	cvCopy(img,img1,bin);
	cvShowImage("原图", img);
	cvShowImage("bin",bin);
	cvShowImage("聚类图像",img1);
	cvWaitKey(0);

得到图像效果如下：

(2)提取标签1

将val改为1，得到图像效果如下：

(3)提取标签2

将val改为2，得到图像效果如下：