-
Notifications
You must be signed in to change notification settings - Fork 0
/
bayes_decision_by_KNN.py
118 lines (93 loc) · 3.25 KB
/
bayes_decision_by_KNN.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
import scipy.io as scio
import numpy as np
import cv2
from get_mask import get_mask
## 距离函数
# 欧氏距离
def euclidean(x, y):
return np.sqrt(np.sum((x - y)**2, axis=1))
# 曼哈顿距离
def manhattan(x, y):
return np.sum(np.abs(x - y), axis=1)
# KNN算法
def KNN(konwed_data, X, k=11, dis_method='e'):
'''
konwed_data: 已知标签的样本特征数据；gray -> (N, 2); rgb -> (N, 4)；第最后一列表示标签
X: 待分类的样本数据; gray -> (n, 1); rgb -> (n, 3)
k: 自定义的邻域范围：选取与待分类点最近的k个已知样本
dis_method: 距离度量函数，默认欧氏距离，指定为'm'变为曼哈顿距离。可选参数['e', 'm']
return:
labels: 返回每个样本点分类后的label,标签-1 or 1; shape -> (n,)
'''
# 计算先验概率
w_1 = np.sum(konwed_data[:, -1] == 1) / konwed_data.shape[0]
w_2 = 1 - w_1
# print(w_1, w_2)
# 设置风险权重
w1 = 1
w2 = 1
labels = np.zeros(X.shape[0])
# 计算距离
for i, x in enumerate(X):
if dis_method == 'e':
# 采用欧氏距离
dis = euclidean(x, konwed_data[:, :-1])
else:
dis = manhattan(x, konwed_data[:, :-1])
idx = np.argsort(dis)
k_nerbor = konwed_data[idx[:k]] # 取距离前k个的样本
count_1 = np.sum(k_nerbor[:, -1] == 1)
count_2 = k - count_1
p_1 = count_1 / k * w_1 * w1
p_2 = count_2 / k * w_2 * w2
if p_1 > p_2:
labels[i] = -1
else:
labels[i] = 1
return labels
# 加载训练样本
array_sample = scio.loadmat('array_sample.mat')
array_sample = array_sample['array_sample']
img_name = '317.bmp'
if '309' in img_name:
# 加载测试样本mask
Mask = scio.loadmat('Mask.mat')
mask = Mask['Mask']
else:
mask = get_mask(img_name)
# 获取测试样本图像
nemo_rgb = cv2.imread(img_name, cv2.IMREAD_COLOR)
nemo_rgb = cv2.cvtColor(nemo_rgb, cv2.COLOR_BGR2RGB)
nemo_gray = cv2.imread(img_name, cv2.IMREAD_GRAYSCALE)
# 获取带标签的灰度样本数据
konwed_data_gray = array_sample[:, [0, 4]]
origin_size_gray = nemo_gray.shape
X_gray = np.reshape(nemo_gray / 255, (-1))
# 进行灰度knn分类
gray_labels = KNN(konwed_data_gray, X_gray)
nemo_gray_out = np.reshape(gray_labels, origin_size_gray) * mask
# 为分类后的Nemo鱼上色
nemo_gray_out[nemo_gray_out == 1] = 100
nemo_gray_out[nemo_gray_out == -1] = 255
nemo_gray_out = np.array(nemo_gray_out, dtype=np.uint8)
# 获取带标签的彩色样本数据
konwed_data_rgb = array_sample[:, 1:]
origin_size_rgb = nemo_rgb.shape
X_rgb = np.reshape(nemo_rgb / 255, (-1, 3))
# 进行彩色knn分类
rgb_labels = KNN(konwed_data_rgb, X_rgb)
# 为分类后的Nemo鱼上色
nemo_rgb_out_2d = np.reshape(rgb_labels, origin_size_rgb[:2]) * mask # (240, 320)
nemo_rgb_out = np.expand_dims(nemo_rgb_out_2d,2).repeat(3,axis=2) # (240, 320, 3)
nemo_rgb_out[nemo_rgb_out_2d == -1] = np.array([255, 0, 0], dtype=np.uint8)
nemo_rgb_out[nemo_rgb_out_2d == 1]= np.array([255, 255, 255], dtype=np.uint8)
nemo_rgb_out[nemo_rgb_out_2d == 0] = np.array([0, 0, 0], dtype=np.uint8)
# 可视化
mask[mask == 1] = 255
cv2.imshow('Nemo_mask', mask)
cv2.imshow('Nemo_gray_origin', nemo_gray)
cv2.imshow('Nemo_rgb_origin', cv2.cvtColor(nemo_rgb, cv2.COLOR_BGR2RGB))
cv2.imshow('Nemo_gray_out', nemo_gray_out)
cv2.imshow('Nemo_rgb_out', nemo_rgb_out)
cv2.waitKey(0)
cv2.destroyAllWindows()