基于支持向量机SVM的人脸识别

 1           
 2 7、代码测试
 3 
 4 #!/usr/bin/python
 5 # encoding: utf-8
 6 
 7 
 8 """
 9 @author : 杜函敏
10 @contact : duhanmin@foxmail.com
11 @File : SVMImplementation.py
12 @time : 2017/7/24 23:14
13 
14 """
15 import numpy as np
16 import pylab as pl
17 from sklearn import svm
18 
19 
20 np.random.seed(0)
21 X = np.r_[np.random.randn(20, 2) - [2, 2], np.random.randn(20, 2) + [2, 2]]
22 Y = [0] * 20 + [1] * 20
23 
24 
25 clf = svm.SVC(kernel=‘linear‘)
26 clf.fit(X, Y)
27 
28 #得到的是几何形式，我们要转化为点斜式方程
29 w = clf.coef_[0]
30 a = -w[0] / w[1]
31 xx = np.linspace(-5, 5)
32 yy = a * xx - (clf.intercept_[0]) / w[1]
33 
34 b = clf.support_vectors_[0]
35 yy_down = a * xx + (b[1] - a * b[0])
36 b = clf.support_vectors_[-1]
37 yy_up = a * xx + (b[1] - a * b[0])
38 
39 
40 print ("w: ", w)
41 print ("a: ", a)
42 print ("support_vectors_: ", clf.support_vectors_)
43 print ("clf.coef_: ", clf.coef_)
44 
45 pl.plot(xx, yy, ‘k-‘)
46 pl.plot(xx, yy_down, ‘k--‘)
47 pl.plot(xx, yy_up, ‘k--‘)
48 
49 pl.scatter(clf.support_vectors_[:, 0], clf.support_vectors_[:, 1],
50 s=80, facecolors=‘none‘)
51 pl.scatter(X[:, 0], X[:, 1], c=Y, cmap=pl.cm.Paired)
52 
53 pl.axis(‘tight‘)
54 pl.show()

  1 #!/usr/bin/python
  2 # encoding: utf-8
  3 
  4 
  5 """
  6 @author : 杜函敏
  7 @contact : duhanmin@foxmail.com
  8 @File : SVMImplementation.py
  9 @time : 2017/7/24 23:14
 10 """
 11 
 12 from __future__ import print_function
 13 
 14 from time import time
 15 import logging
 16 import matplotlib.pyplot as plt
 17 
 18 from sklearn.model_selection import train_test_split
 19 from sklearn.datasets import fetch_lfw_people
 20 from sklearn.model_selection import GridSearchCV
 21 from sklearn.metrics import classification_report
 22 from sklearn.metrics import confusion_matrix
 23 from sklearn.decomposition import PCA
 24 from sklearn.svm import SVC
 25 
 26 plt.rcParams[‘font.sans-serif‘] = [‘SimHei‘] #指定默认字体
 27 plt.rcParams[‘axes.unicode_minus‘] = False #解决保存图像是负号‘-‘显示为方块的问题
 28 
 29 
 30 #在控制台上显示日志
 31 logging.basicConfig(level=logging.INFO, format=‘%(asctime)s %(message)s‘)
 32 
 33 ###############################################################################
 34 # 下载数据，如果尚未在磁盘上，并将其作为numpy数组加载
 35 
 36 lfw_people = fetch_lfw_people(min_faces_per_person=70, resize=0.4)
 37 
 38 # shape求图片矩阵的维度大小
 39 # n_samples图数量 h,w图的大小
 40 n_samples, h, w = lfw_people.images.shape
 41 
 42 # data中为每个图片矩阵的特征向量（列）
 43 X = lfw_people.data
 44 n_features = X.shape[1]
 45 
 46 # 提取不同人的身份标记
 47 y = lfw_people.target
 48 
 49 # 提取人的名字
 50 target_names = lfw_people.target_names
 51 n_classes = target_names.shape[0]
 52 
 53 print("*************总数据集大小**********************")
 54 print("**图数量: %d" % n_samples)
 55 print("**特征向量数: %d" % n_features)
 56 print("**人数: %d" % n_classes)
 57 print("*************总数据集大小**********************")
 58 
 59 
 60 ###############################################################################
 61 # 分训练和测试集
 62 X_train, X_test, y_train, y_test = train_test_split(
 63 X, y, test_size=0.25)
 64 ###############################################################################
 65 # 特征提取/降维
 66 n_components = 150
 67 print("从 %d 个维度中提取到 %d 维度" % (X_train.shape[0],n_components))
 68 #主成分分析建模
 69 pca = PCA(n_components=n_components, whiten=True).fit(X_train)
 70 eigenfaces = pca.components_.reshape((n_components, h, w))
 71 print("根据主成分进行降维开始")
 72 X_train_pca = pca.transform(X_train)
 73 X_test_pca = pca.transform(X_test)
 74 print("降维结束")
 75 ###############################################################################
 76 # 训练SVM
 77 print("训练SVM分类模型开始")
 78 t0 = time()
 79 #构建归类精确度5x6=30
 80 param_grid = {‘C‘: [1e3, 5e3, 1e4, 5e4, 1e5],
 81 ‘gamma‘: [0.0001, 0.0005, 0.001, 0.005, 0.01, 0.1], }
 82 #图片用rbf核函数,权重自动选取
 83 clf = GridSearchCV(SVC(kernel=‘rbf‘, class_weight=‘balanced‘), param_grid)
 84 clf = clf.fit(X_train_pca, y_train)
 85 print("SVM训练结束，结果如下：" "SVM训练用时 %0.3fs" % (time() - t0))
 86 print(clf.best_estimator_)
 87 
 88 # ###############################################################################
 89 print("测试集SVM分类模型开始")
 90 
 91 t0 = time()
 92 y_pred = clf.predict(X_test_pca)
 93 print("测试集用时 %0.3fs" % (time() - t0))
 94 
 95 print("误差衡量")
 96 # 数据中1的个数为a，预测1的次数为b，预测1命中的次数为c
 97 # 准确率 precision = c / b
 98 # 召回率 recall = c / a
 99 # f1_score = 2 * precision * recall / (precision + recall)
100 
101 print(classification_report(y_test, y_pred, target_names=target_names))
102 print("预测值和实际值对角矩阵")
103 print(confusion_matrix(y_test, y_pred, labels=range(n_classes)))
104 
105 
106 # ###############################################################################
107 # 画图
108 def plot_gallery(images, titles, h, w, n_row=3, n_col=4):
109 plt.figure(figsize=(1.8 * n_col, 2.4 * n_row))
110 plt.subplots_adjust(bottom=0, left=.01, right=.99, top=.90, hspace=.35)
111 for i in range(n_row * n_col):
112 plt.subplot(n_row, n_col, i + 1)
113 plt.imshow(images[i].reshape((h, w)), cmap=plt.cm.gray)
114 plt.title(titles[i], size=12)
115 plt.xticks(())
116 plt.yticks(())
117 
118 # 绘制一部分测试集上的预测结果
119 def title(y_pred, y_test, target_names, i):
120 ##以空格为分隔符，把y_pred分成一个list。分割的次数1。[-1]取最后一个
121 pred_name = target_names[y_pred[i]].rsplit(‘ ‘, 1)[-1]
122 true_name = target_names[y_test[i]].rsplit(‘ ‘, 1)[-1]
123 return ‘预测值: %s\n 实际值: %s‘ % (pred_name, true_name)
124 
125 prediction_titles = [title(y_pred, y_test, target_names, i)
126 for i in range(y_pred.shape[0])]
127 
128 plot_gallery(X_test, prediction_titles, h, w)
129 
130 # 画人脸咯，eigenfaces主成分特征脸
131 eigenface_titles = ["特征脸 %d" % i for i in range(eigenfaces.shape[0])]
132 plot_gallery(eigenfaces, eigenface_titles, h, w)
133 
134 plt.show()