python數(shù)據(jù)處理 根據(jù)顏色對圖片進(jìn)行分類的方法

import os
import numpy as np
import skimage
import matplotlib.pyplot as plt
from skimage import io 				#讀取圖片
from skimage import exposure		#調(diào)用調(diào)對比度的方法	rescale_intensity、equalize_hist
from skimage.filters import gaussian	#高斯
from skimage import img_as_float  #圖片unit8類型到float
from scipy.cluster.vq import kmeans,vq,whiten  #聚類算法
import shutil	#文件夾內(nèi)容刪除
 
class Path(object):
	def __init__(self):
		self.path = r"D:\PYscrapy\get_lixiaoran\picture"
		self.pathlist = []	#原始圖片列表
		self.page = 0
 
	def append(self):					#將每張圖片的路徑加載到列表中
		much = os.listdir(self.path)
		for i in range(len(much)):
			repath = os.path.join(self.path,str(self.page)+'.jpg')
			self.page +=1
			self.pathlist.append(repath)
		return self.pathlist
 
class Contrast(object):
	def __init__(self,pathlist):
		self.pathlist = pathlist
		self.contrastlist = []	#改變對比度之后的圖片列表
		self.path3 = r"D:\PYscrapy\get_lixiaoran\picture2"
		self.page2 = 0
 
	def balance(self):			#將每張圖片進(jìn)行對比度的處理，兩種方式 1：均衡化 2：從某個值開始取極值
		if os.path.exists(self.path3) == False:
			os.mkdir(self.path3)
 
		# for lis in self.pathlist:
		# 	data = skimage.io.imread(lis)
		# 	equalized = exposure.equalize_hist(data)	#方法一這里使用個人人為更好的均衡化處理對比度的方法
		# 	self.contrastlist.append(equalized)
 
		for lis in self.pathlist:
			data = skimage.io.imread(lis)
			high_contrast = exposure.rescale_intensity(data,in_range=(20,220))	#方法二 以20和220取兩端極值
			self.contrastlist.append(high_contrast)
 
		for img in self.contrastlist:
			repath = os.path.join(self.path3,str(self.page2)+'.jpg')		#保存修改后的圖片
			skimage.io.imsave(repath,img)
			self.page2 +=1
 
class Filter(Contrast):
	def __init__(self,pathlist):
		super().__init__(pathlist)
		self.path41 = self.path3
		self.path42 = r"D:\PYscrapy\get_lixiaoran\picture3"
		self.page3 = 0
		self.filterlist = []
 
	def filte_r(self):
		img = os.listdir(self.path41)	#讀取文件內(nèi)容
		if os.path.exists(self.path42) == False:
			os.mkdir(self.path42)
		for lis in range(len(img)):			#循環(huán)做每張圖片的高斯過濾
			path = os.path.join(self.path41,str(lis)+r'.jpg')
			img = skimage.io.imread(path)
			gas = gaussian(img,sigma=3)		#multichannel=False	去掉顏色2D
			self.filterlist.append(gas)
			path_gas = os.path.join(self.path42,str(self.page3)+r'.jpg')
			skimage.io.imsave(path_gas,gas)
			self.page3 +=1
		return self.path42
 
class Vectoring(object):
	def __init__(self,filter_path):
		self.path51 = filter_path
		self.diff = []
		self.calculate = []
 
	def vector(self):
		numbers = os.listdir(self.path51)	#獲取文件夾內(nèi)容
		os.chdir(self.path51)		#切換路徑
		for i in range(len(numbers)):
			self.diff.append([])
			for j in range(4):
				self.diff[i].append([])		#diff[[number],[img_float],[bin_centers],[hist]]
 
		for cnt,number in enumerate(numbers):
			img_float = img_as_float(skimage.io.imread(number))		#將圖像ndarry nint8->float
			hist,bin_centers = exposure.histogram(img_float,nbins=10)	#取圖像的	每個區(qū)間的像素值	分隔區(qū)間
			self.diff[cnt][0] = number
			self.diff[cnt][1] = img_float
			self.diff[cnt][2] = bin_centers	#把數(shù)據(jù)添加到diff中
			self.diff[cnt][3] = hist
 
		for i,j in enumerate(self.diff):		#使用hist和bin_centers相乘來降維,向量化
			self.calculate.append([y*self.diff[i][3][x] for x,y in enumerate(self.diff[i][2])])	#這里可能需要理解一下，就是涉及的參數(shù)有點(diǎn)多
		for i in range(len(self.diff)):
			self.diff[i].append(self.calculate[i])	#將特征向量calculate也加入到diff中
 
		return self.diff 			#diff[[number],[img_float],[bin_centers],[hist],[calculate]]
 
class Modeling(Vectoring):
	def __init__(self,filter_path,K):
		super().__init__(filter_path)
		self.K = K
 
	def model(self):
		diff = self.vector()
		calculate = []
		for i in range(len(diff)):
			calculate.append(diff[i][4])
		spot = whiten(calculate)			#這里使用scipy的k-means方法來對圖片進(jìn)行分類
		center,_ = kmeans(spot,self.K)		#如果對scipy的k-means不熟悉，前面有專門的講解
		cluster,_ = vq(spot,center)
		return diff,cluster 	#獲得預(yù)測值
		
class Predicting(object):
	def __init__(self,predicted_diff,predicted_cluster,K):
		self.diff = predicted_diff
		self.cluster = predicted_cluster
		self.path52 = r'D:\PYscrapy\get_lixiaoran\picture4'
		self.K = K
 
	def predicted(self):
		if os.path.exists(self.path52) == True:
			much = shutil.rmtree(self.path52)
			os.mkdir(self.path52)
		else:
			os.mkdir(self.path52)
		os.chdir(self.path52)
		for i in range(self.K):			#創(chuàng)建K個文件夾
			os.mkdir('classify{}'.format(i))
		for i,j in enumerate(self.cluster):
			skimage.io.imsave('classify{}\\{}'.format(j,self.diff[i][0]),self.diff[i][1])	#根據(jù)圖片的分類來將它們保存至對應(yīng)的文件夾
 
if __name__=="__main__":
	np.random.seed(10)
	#文件路徑添加
	start = Path()
	pathlist = start.append()
 
	#對比度類
	second = Contrast(pathlist)
	second.balance()	#get改變對比度后的圖片個數(shù)
 
	#高斯過濾
	filte = Filter(pathlist)
	filter_path = filte.filte_r()
 
	#數(shù)據(jù)提取及向量化
	vectoring = Vectoring(filter_path)
 
	#K值的自定義
	K = 3
 
	#建模
	modeling = Modeling(filter_path,K)
	predicted_diff,predicted_cluster = modeling.model()
 
	#預(yù)測
	predicted = Predicting(predicted_diff,predicted_cluster,K)
	predicted.predicted()