基于Python实现图像分割算法的（源码+文档）

#基于遗传算法的图像分割 #目标：最大化目标与背景之间的类间方差或最小化目标像素内的类间方差 #编码：8位二进制编码 #适应度函数： #选择：轮盘赌 #变异：单点变异 import numpy as np import cv2 import random import matplotlib.pyplot as plt from skimage import transform,data #读入灰度图像 Image = cv2.imread('test_3.jpeg',cv2.IMREAD_GRAYSCALE) height, width = Image.shape[:2] Image_re = cv2.resize(Image,(int(height*0.1),int(width*0.1))) im = np.array(Image_re,dtype=np.float) im_copy = np.array(Image,dtype=np.float) im_res = im_copy.copy() im_ga = np.copy(im) lchorm = 8 #染色体长度 popsize = 10 #种群大小 cross_rate = 0.7 #交叉概率 mutation_rate = 0.4 #变异概率 maxgen = 150 #最大代数 MN = np.shape(im_ga) fit = [] yuzhi = [] b = [] fitness = [] popsize1 = 0 oldpop1 = [] fitness1 = [] b1 = [] #初始化种群 def Initpop(): global lchorm,popsize,im_ga oldpop = [] for i in range(popsize): temp = '0'*lchorm for j in range(lchorm): if random.random()<0.5: temp = temp[0:j]+'1'+temp[j+1:] oldpop.append(temp) return oldpop #a = Initpop(lchorm,popsize,im_ga) #print(a) oldpop = Initpop() #计算适应度并排序 def Fitness_order(): global gen,popsize,oldpop,MN,lchorm,fit,yuzhi,popsize1,oldpop1,fitness,b,fitness1,b1 if popsize>=5: popsize = round(popsize-0.03*gen) #到进化末期的时候调整种群规模和交叉、变异概率 if gen == 75: cross_rate = 0.3 mutation_rate = 0.3 if gen > 1: t = oldpop j = popsize1 for i in range(popsize): if j >= 1: oldpop[i] = t[j-1] j = j-1 #计算适应度值并排序 fitness = [] b = [] for i in range(popsize): lowsum = 0.0 higsum = 0.0 lownum = 0.0 hignum = 0.0 temp_ = oldpop[i] c = int(temp_,2) b.append(c*255/(2**lchorm-1)) for x in range(MN[0]): for y in range(MN[1]): if(im_ga[x,y] <= b[i]): lowsum = lowsum + im_ga[x,y]#统计低于阈值的灰度值的总和 lownum = lownum + 1.0 #统计低于阈值的灰度值的像素的总个数 else: higsum = higsum + im_ga[x,y]#统计高于阈值的灰度值的总和 hignum = hignum + 1.0 #统计高于阈值的灰度值的像素的总个数 #求u1、u2为对应于两类的平均灰度值 if lownum != 0: u1 = lowsum/lownum else: u1 = 0 if hignum != 0: u2 = higsum/hignum else: u2 = 0 x = lownum*hignum*(u1-u2)**2 fitness.append(x)#计算适应度 test = 1 if gen == 1: #如果为第一代，从小往大排序 for i in range(popsize): j = i + 1 while(j <=popsize-1): if fitness[i] > fitness[j]: tempf = fitness[i] tempc = oldpop[i] tempb = b[i] b[i] = b[j] b[j] = tempb fitness[i] = fitness[j] fitness[j] = tempf oldpop[i] = oldpop[j] oldpop[j] = tempc j = j+1 fitness1 = fitness.copy() b1 = b.copy() oldpop1 = oldpop.copy() popsize1 = popsize else: for i in range(popsize): j = i + 1 while(j <=popsize-1): if fitness[i] > fitness[j]: tempf = fitness[i] tempc = oldpop[i] tempb = b[i] b[i] = b[j] b[j] = tempb fitness[i] = fitness[j] fitness[j] = tempf oldpop[i] = oldpop[j] oldpop[j] = tempc j = j+1 for i in range(popsize1): j = i + 1 while(j <= popsize1-1): if fitness1[i] > fitness1[j]: tempf = fitness1[i] tempc = oldpop1[i] tempb = b1[i] b1[i] = b1[j] b1[j] = tempb fitness1[i] = fitness1[j] fitness1[j] = tempf oldpop1[i] = oldpop1[j] oldpop1[j] = tempc j = j+1 if gen == 1: fit.append(fitness[-1]) yuzhi.append(b[-1]) yuzhisum = 0 else: if (fitness[-1]>fitness1[-1]): yuzhi.append(b[-1]) fit.append(fitness[-1]) else: yuzhi.append(b1[-1]) fit.append(fitness1[-1]) return def Select(): global gen,fitness,popsize,oldpop,b,fitness1,popsize1,oldpop1,b1 s = popsize1 + 1 for j in range(popsize-1,-1,-1): if fitness[-1]<fitness1[j]: s = j temp=oldpop.copy() if s!=popsize1 + 1: if gen < 50:#小于50代用上一代中用适应度值大于当前代的个体随机代替当前代中的个体 for i in range(s,popsize1): j = int(random.random()*popsize-0.00001) temp[j] = oldpop1[i] b[j] = b1[i] fitness[j] = fitness1[i] else: if gen < 100: j = 0 for i in range(s,popsize1): temp[j] = oldpop1[i] b[j] = b1[i] fitness[j] = fitness1[i] j = j+1 else: j = popsize1-1 for i in range(int(popsize/2)): temp[i] = oldpop1[j] b[i] = b1[j] fitness[i] = fitness1[j] j = j-1 b1 = b.copy() fitness1 = fitness.copy() oldpop1 = temp.copy() popsize1 = popsize return temp def Crossover(): global temp,popsize,cross_rate,lchorm parent = [] a = [] for i in range(popsize): if (random.random()<cross_rate): parent.append(temp[i]) a.append(i) j = len(parent) if j%2 != 0: j = j-1 if j >= 2: for k in range(0,j,2): cutpoint = random.randint(0,lchorm-1) f = k temp[a[f]] = parent[f][:cutpoint]+parent[f+1][cutpoint:] temp[a[f+1]] = parent[f+1][:cutpoint]+parent[f][cutpoint:] return def Mutation(): global popsize,lchorm,mutation_rate,temp,newpop,oldpop gasum = lchorm*popsize mutnum = round(mutation_rate*gasum) for i in range(mutnum): s = random.randint(0,lchorm-1) t = random.randint(0,popsize-1) if temp[t][s] == '1': temp[t] = temp[t][:s]+'0'+temp[t][s+1:] else: temp[t] = temp[t][:s]+'1'+temp[t][s+1:] oldpop = temp.copy() return for gen in range(maxgen): gen = gen+1 Fitness_order() temp = Select() Crossover() Mutation() for i in range(np.shape(im_res)[0]): for j in range(np.shape(im_res)[1]): if im_res[i][j] <= yuzhi[-1]: im_res[i][j] = 0 else: im_res[i][j] = 255 fig, axes = plt.subplots(1,2, figsize=(10, 6)) #axes[0] = plt.plot(range(gen),fit) #print(fit) #print(yuzhi) axes[0].imshow(im_copy, cmap='gray') axes[0].set_title('first') axes[0].set_axis_off() axes[1].imshow(im_res, cmap='gray') axes[1].set_title('second') axes[1].set_axis_off() #axes[1][0].plot(range(gen),fit) print(yuzhi[-1]) plt.show()