存在需要解决的问题:
1、kmean是随机聚类的,当设置K值后,每次聚类完后分配的标签也是随机的,如果直接根据聚类后的标签给着色,客户每一次聚类后发现颜色都跟上一次不一样,又开始哔哩哔哩了;
2、可通过设置kmean的参数cv2.KMEANS_USE_INITIAL_LABELS,可一定程度上降低每次运行后标签变动太大,但不能彻底解决这个问题;
3、可通过改变kmean聚类的标签来实现,每次在标签图上着色的颜色位置都是固定的。
简单的方案:(1)对聚类标签的数量进行统计,按占比例大小进行从小到大排序;
(2)对原聚类的标签使用排序后的索引进行强制打标签;
存在问题:每次运行kmean聚类后,着色颜色基本上能稳定,但是还是偶尔会变。
numpy获取数组下标:
index = np.argsort(ratio)
numpy根据两组数据创建字典:
dic = dict(zip(rt,ratio))
import cv2
import numpy as np
import pydensecrf.densecrf as dcrf
try:
from cv2 import imread, imwrite
except ImportError:
# 如果没有安装OpenCV,就是用skimage
from skimage.io import imread, imsave
imwrite = imsave
from pydensecrf.utils import unary_from_labels, create_pairwise_bilateral, create_pairwise_gaussian
blue = [255, 20, 20] #蓝色
green = [50, 255, 50] #绿色
red = [50, 50, 255] #红色
light_green = [255, 255, 128] #青色
white = [50, 255, 255] #橙色
#颜色字典
color_dit = np.array([blue, green, red, light_green, white])
#标签着色函数
def decode_segmap(predit_map,color_map,num_class):
img = predit_map[:, :, 0] if len(predit_map.shape) == 3 else predit_map
img_out = np.zeros(img.shape +(3,))
if num_class > len(color_map):
num_class = len(color_map)
for i in range(num_class):
img_out[img == i,:] = color_dit[i]
return img_out
def CRFs(original_image_path, predicted_image_path, CRF_image_path):
img = original_image_path
# 将predicted_image的RGB颜色转换为uint32颜色 0xbbggrr
anno_rgb = predicted_image_path.astype(np.uint32)
anno_lbl = anno_rgb[:, :, 0] + (anno_rgb[:, :, 1] << 8) + (anno_rgb[:, :, 2] << 16)
# 将uint32颜色转换为1,2,...
colors, labels = np.unique(anno_lbl, return_inverse=True)
# 如果你的predicted_image里的黑色(0值)不是待分类类别,表示不确定区域,即将分为其他类别
# 那么就取消注释以下代码
# HAS_UNK = 0 in colors
# if HAS_UNK:
# colors = colors[1:]
# 创建从predicted_image到32位整数颜色的映射。
colorize = np.empty((len(colors), 3), np.uint8)
colorize[:, 0] = (colors & 0x0000FF)
colorize[:, 1] = (colors & 0x00FF00) >> 8
colorize[:, 2] = (colors & 0xFF0000) >> 16
# 计算predicted_image中的类数。
n_labels = len(set(labels.flat))
# n_labels = len(set(labels.flat)) - int(HAS_UNK) ##如果有不确定区域,用这一行代码替换上一行
###########################
### 设置CRF模型 ###
###########################
use_2d = False
# use_2d = True
###########################################################
##不是很清楚什么情况用2D
##作者说“对于图像,使用此库的最简单方法是使用DenseCRF2D类”
##作者还说“DenseCRF类可用于通用(非二维)密集CRF”
##但是根据我的测试结果一般情况用DenseCRF比较对
#########################################################33
if use_2d:
# 使用densecrf2d类
d = dcrf.DenseCRF2D(img.shape[1], img.shape[0], n_labels)
# 得到一元势(负对数概率)
U = unary_from_labels(labels, n_labels, gt_prob=0.2, zero_unsure=None)
# U = unary_from_labels(labels, n_labels, gt_prob=0.2, zero_unsure=HAS_UNK)## 如果有不确定区域,用这一行代码替换上一行
d.setUnaryEnergy(U)
# 增加了与颜色无关的术语,功能只是位置而已
d.addPairwiseGaussian(sxy=(3, 3), compat=3, kernel=dcrf.DIAG_KERNEL,
normalization=dcrf.NORMALIZE_SYMMETRIC)
# 增加了颜色相关术语,即特征是(x,y,r,g,b)
d.addPairwiseBilateral(sxy=(80, 80), srgb=(13, 13, 13), rgbim=img, compat=10,
kernel=dcrf.DIAG_KERNEL,
normalization=dcrf.NORMALIZE_SYMMETRIC)
else:
# 使用densecrf类
d = dcrf.DenseCRF(img.shape[1] * img.shape[0], n_labels)
# 得到一元势(负对数概率)
U = unary_from_labels(labels, n_labels, gt_prob=0.7, zero_unsure=None)
# U = unary_from_labels(labels, n_labels, gt_prob=0.7, zero_unsure=HAS_UNK)## 如果有不确定区域,用这一行代码替换上一行
d.setUnaryEnergy(U)
# 这将创建与颜色无关的功能,然后将它们添加到CRF中
feats = create_pairwise_gaussian(sdims=(3, 3), shape=img.shape[:2])
d.addPairwiseEnergy(feats, compat=3, kernel=dcrf.DIAG_KERNEL,
normalization=dcrf.NORMALIZE_SYMMETRIC)
# 这将创建与颜色相关的功能,然后将它们添加到CRF中
feats = create_pairwise_bilateral(sdims=(80, 80), schan=(13, 13, 13),
img=img, chdim=2)
d.addPairwiseEnergy(feats, compat=10,
kernel=dcrf.DIAG_KERNEL,
normalization=dcrf.NORMALIZE_SYMMETRIC)
####################################
### 做推理和计算 ###
####################################
# 进行5次推理
Q = d.inference(20)
# 找出每个像素最可能的类
MAP = np.argmax(Q, axis=0)
# 将predicted_image转换回相应的颜色并保存图像
MAP = colorize[MAP, :]
#imwrite(CRF_image_path, MAP.reshape(img.shape))
return MAP.reshape(img.shape)
''''
算法调用函数入库
参数image,表示需要进行聚类的图像
参数k,表示keams的K值,默认大小为5
'''
def segment_kmeans_color_(image,k=5):
height, weight,_ = image.shape
image_copy = image.copy()
# 3个通道展平
img_flat = image_copy.reshape((image.shape[0] * image.shape[1], 3))
img_flat = np.float32(img_flat)
# kmean聚类参数
criteria = (cv2.TERM_CRITERIA_EPS + cv2.TermCriteria_MAX_ITER, 200, 0.5)
flags = cv2.KMEANS_USE_INITIAL_LABELS
# 聚类
compactness, labels, centers = cv2.kmeans(img_flat, k, None, criteria, 20, flags)
#把聚类标签还原为原图的shape
labels = labels.reshape(height, weight)
#获取聚类族群的个数
cluster = centers.shape[0]
#统计聚类后每一类占整个图像像素的比例值
ratio = []
for i in range(cluster):
ratio.append(len(labels[labels==i]) / (height * weight))
#获取从小到大排列的比例值在ratio中的索引值,即标签值
index = np.argsort(ratio)
#由于kmean聚类后分配标签并不是按照类别在原图所占比例大小进行分配的,所以这里进行标签更正
#以聚类后所占原图比例进行从小到大分配标签
correct_label = np.zeros(labels.shape)
for i in range(cluster):
correct_label[labels==index[i]]=i
#构造字典
rt = []
for k in range(cluster):
rt.append('ratio' + str(k))
ratio.sort()
dic = dict(zip(rt,ratio))
#对标签进行着色
color_labels = decode_segmap(correct_label,centers, centers.shape[0])
#使用条件随机场对分割进行后处理
color_labels = CRFs(image,color_labels,None)
#把分割彩色图和原图进行融合
image_res = cv2.addWeighted(image, 1, color_labels, 0.35, 0)
cv2.imshow("image", image_res)
dic['image'] = image_res
return dic
if __name__ == '__main__':
for i in range(5):
image = cv2.imread('G:\\Data\\segment\\20200529165615.png', cv2.IMREAD_COLOR)
k =5
res = segment_kmeans_color_(image,k)
cv2.waitKey(0)