-
Notifications
You must be signed in to change notification settings - Fork 3
/
som.py
172 lines (147 loc) · 6.1 KB
/
som.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
# -*- coding: utf-8 -*-
from random import *
from math import *
from PIL import Image
from time import *
#----------- SOM节点 -------------
class Node:
# 初始化
def __init__(self, X=0, Y=0):
self.R=randint(0,255) # R值
self.G=randint(0,255) # G值
self.B=randint(0,255) # B值
self.X=X # X坐标
self.Y=Y # Y坐标
#----------- SOM结构 -------------
class SOM:
# 初始化
def __init__(self, height=500, width=500):
self.height=height # 高度
self.width=width # 宽度
self.radius=height/2+width/2 # 半径
self.total=self.height*self.width # 总数
self.nodes=[[0 for x in range(self.height)] for y in range(self.width)] # 节点结构
# 初始化节点结构
for i in range(self.width):
for j in range(self.height):
self.nodes[i][j]=Node(i,j)
# 计算像素距离
def W_distance(self, r_sample, node):
return sqrt((r_sample[0]-node.R)**2+(r_sample[1]-node.G)**2+(r_sample[2]-node.B)**2)
# 计算位置距离
def distance(self, node1, node2):
return sqrt((node1.X-node2.X)**2+(node1.Y-node2.Y)**2)
# 获取最优竞争者
def get_bmu(self,r_sample):
# 竞争匹配数目
match_amt=0
# 最大距离
max_dist=1000000.0
# 匹配列表
match_list=[]
# 遍历节点
for i in range(self.width):
for j in range(self.height):
t_dist=self.W_distance(r_sample,self.nodes[i][j])
# 更新新距离
if t_dist<max_dist:
max_dist=t_dist
match_list=[]
match_list.append(self.nodes[i][j])
match_amt=1
# 添加节点
elif t_dist==max_dist:
match_list.append(self.nodes[i][j])
match_amt+=1
# 返回优胜节点
return match_list[randint(0,match_amt-1)]
# 更新领域值
def scale_neighbors(self, bmu_loc, r_sample, times):
# 计算领域范围半径
R2=(int)(((float)(self.radius)*(1.0-times)/2.0))+1
# 计算节点距离
outer=Node(R2,R2)
center=Node(0,0)
d_normalize=self.distance(center,outer)
print u"radius:",R2,u"d_normalize:",d_normalize
for i in range(-R2,R2+1):
for j in range(-R2,R2+1):
# 半径范围内的节点
if j+bmu_loc.Y >= 0 and j+bmu_loc.Y < self.height and i+bmu_loc.X >= 0 and i+bmu_loc.X < self.width:
# 计算到中心点的距离并归一化
outer=Node(i,j)
distance=self.distance(center,outer)
distance/= d_normalize
# 度量学习率
t=(float)(exp(-1.0*distance**2/0.15))
# 计算随迭代次数减少的程度
t/=(times*4.0+1.0)
# 更新值
self.nodes[i+bmu_loc.X][j+bmu_loc.Y].R = int(self.nodes[i+bmu_loc.X][j+bmu_loc.Y].R*(1-t) + r_sample[0]*t)
self.nodes[i+bmu_loc.X][j+bmu_loc.Y].G = int(self.nodes[i+bmu_loc.X][j+bmu_loc.Y].G*(1-t) + r_sample[1]*t)
self.nodes[i+bmu_loc.X][j+bmu_loc.Y].B = int(self.nodes[i+bmu_loc.X][j+bmu_loc.Y].B*(1-t) + r_sample[2]*t)
# 输出图像
# self.save_image(str((int)(100*times)))
self.save_image("result_2")
sleep(1)
# 保存图像
def save_image(self, imagename):
newImage = Image.new ("RGBA", (self.width,self.height), (0,0,0))
newpix = newImage.load()
for i in range(self.width):
for j in range(self.height):
newpix[i,j] = (self.nodes[i][j].R,self.nodes[i][j].G,self.nodes[i][j].B,255)
newImage.save("data\\result_image\\"+imagename+".jpg")
#----------- 程序的入口处 -----------
if __name__ == "__main__":
print u"""
--------------------------------------------------------
程序：SOM图片生成程序
作者：DiamonJoy
日期：2015-11-24
语言：Python 2.7
--------------------------------------------------------
"""
print u'请按下回车开始程序'
raw_input(' ')
print u"1.读入测试图片中..."
# 测试图片
myimage = Image.open('data\\source_image\\2.jpg')
# 图片尺寸
myimage_with = myimage.size[0]
myimage_height = myimage.size[1]
print u"image_with:",myimage_with,u"image_height:",myimage_height
print u""
# 读入图片像素值
mypix = myimage.load()
print u"2.训练中..."
# Som训练
mySom = SOM(myimage_with,myimage_height)
# 当前迭代次数，对于邻域和学习率有影响
times=0.0
# 最大迭代次数
MAX_ITER=100
# 次数增加幅度
T_INC=1.0/(float)(MAX_ITER)
# print T_INC
# 循环迭代
while(True):
if times<1.0:
# 获取随机样本
random_x=randint(0,myimage_with-1)
random_y=randint(0,myimage_height-1)
r_sample=mypix[random_x,random_y]
# 获取最优竞争者
bmu_loc=mySom.get_bmu(r_sample)
print u"-----------"+str(int(100*times))+u"%"+u"-----------"
print u"sample.x:",random_x,u"sample.y:",random_y,u"sample.R:",r_sample[0],u"sample.G:",r_sample[1],u"sample.B:",r_sample[2]
print u"bmu.x:",bmu_loc.X,u"bmu.y:",bmu_loc.Y,u"bmu.R:",bmu_loc.R,u"bmu.G:",bmu_loc.G,u"bmu.B:",bmu_loc.B
# 更新领域值
mySom.scale_neighbors(bmu_loc,r_sample,times)
print u""
# 更新当前迭代次数
times+=T_INC
else:
print u"-----------"u"100%"+u"-----------"
print u"程序结束"
break