使用 k-means 进行图像分割

Question

我正在尝试将 k-means 算法用于图像分割任务。问题是我的程序没有分割图像。你能帮我找出代码中的错误吗？

事实上，我已经将簇数固定为32。我使用了以下数据结构：

3 个数组 bleu,vert,rouge 来存储每个像素的 RGB 值
3 个数组 cluster_bleu,cluster_rouge,cluster_vert 来存储每个簇的 RGB 值

groupe[i,0]=k 将每个像素 i 映射到簇 k

import cv2

import numpy

import random
def main():
    MAX_LARGEUR = 400
    MAX_HAUTEUR = 400

    K = 32 #Le fameux parametre K de l'algorithme
    imagecolor = cv2.imread('perr.jpg')


    if imagecolor.shape[0] > MAX_LARGEUR or imagecolor.shape[1] > MAX_HAUTEUR:
        factor1 = float(MAX_LARGEUR) / imagecolor.shape[0]
        factor2 = float(MAX_HAUTEUR) / imagecolor.shape[1]
        factor = min(factor1, factor2)
        imagecolor = cv2.resize(imagecolor, None, fx=factor, fy=factor, interpolation=cv2.INTER_AREA)


    nb_pixels = imagecolor.shape[0] * imagecolor.shape[1]

    bleu = imagecolor[:, :, 0].reshape(nb_pixels, 1)
    vert = imagecolor[:, :, 1].reshape(nb_pixels, 1)
    rouge = imagecolor[:, :, 2].reshape(nb_pixels, 1)


    cluster_bleu = numpy.zeros(K)
    cluster_vert = numpy.zeros(K)
    cluster_rouge = numpy.zeros(K)

    groupe = numpy.zeros((nb_pixels, 1)) 

    for i in range(0,K):
        groupe[i,0]=i

    for i in range(K,nb_pixels):
        groupe[i,0]=random.randint(0, K-1)


    condition =False




    def etape1(indices,i):
    s=indices.size
    rouge_s=0
    vert_s=0
    bleu_s=0
    #calcul de barycentre des points
    if s==0:
        cluster_rouge[i]=0  
        cluster_vert[i]=0
        cluster_bleu[i]=0


    if s >=1:
        for j in range(0,s):
            rouge_s=rouge_s+rouge[indices[j]]
            vert_s=vert_s+vert[indices[j]]
            bleu_s=bleu_s+bleu[indices[j]]

        #mise  jour des clusters 

        cluster_rouge[i]=rouge_s/s  
        cluster_vert[i]=vert_s/s
        cluster_bleu[i]=bleu_s/s        


    iteration=0
    oldGroupe = numpy.copy(groupe)
    while(condition==False) :


    for i in range(0,K):

        indices=numpy.where(groupe==i)[0]
        etape1(indices,i)



    for i in range(0,nb_pixels):
        minimum=10000;
        dist=0;
        index=-1;
        for j in range(0,K):
             dist=(cluster_rouge[j]-rouge[i])**2+(cluster_vert[j]-vert[i])**2+(cluster_bleu[j]-bleu[i])**2;
             if(dist<=minimum):
                minimum=dist;
                index=j;



        groupe[i,0]=index;



    condition=numpy.all(groupe==oldGroupe)

    oldGroupe = numpy.copy(groupe)  


    groupe=numpy.reshape(groupe, (imagecolor.shape[0], imagecolor.shape[1]))




    for i in range(0, imagecolor.shape[0]):
        for j in range(0, imagecolor.shape[1]):
            imagecolor[i,j,0] = (cluster_bleu[groupe[i,j]])
            imagecolor[i,j,1] = (cluster_vert[groupe[i,j]])
            imagecolor[i,j,2] = (cluster_rouge[groupe[i,j]])

    cv2.namedWindow("sortie")
    cv2.imshow("sortie", imagecolor)
    key = cv2.waitKey(0)
if __name__ == "__main__":
    main()

Answer 1

问题是赋值 oldGroupe=groupe; 不复制数组，而是创建具有不同名称 (oldGroupe) 的引用，指向与 groupe 相同的数据。因此，当您更改 groupe 时，您也会更改 oldGroupe，并且 condition 始终为 True。

您想要的是用 oldGroupe = numpy.copy(groupe).

在 groupe 中创建数据副本

使用 k-means 进行图像分割

Image segmentation using k-means

machine-learning

image-processing

computer-vision

k-means

python-2.7