gpt4 book ai didi

Python构建图像分类识别器的方法

转载 作者:qq735679552 更新时间:2022-09-28 22:32:09 24 4
gpt4 key购买 nike

CFSDN坚持开源创造价值,我们致力于搭建一个资源共享平台,让每一个IT人在这里找到属于你的精彩世界.

这篇CFSDN的博客文章Python构建图像分类识别器的方法由作者收集整理,如果你对这篇文章有兴趣,记得点赞哟.

机器学习用在图像识别是非常有趣的话题.

我们可以利用OpenCV强大的功能结合机器学习算法实现图像识别系统.

首先,输入若干图像,加入分类标记。利用向量量化方法将特征点进行聚类,并得出中心点,这些中心点就是视觉码本的元素.

其次,利用图像分类器将图像分到已知的类别中,ERF(极端随机森林)算法非常流行,因为ERF具有较快的速度和比较精确的准确度。我们利用决策树进行正确决策.

最后,利用训练好的ERF模型后,创建目标识别器,可以识别未知图像的内容.

当然,这只是雏形,存在很多问题:

界面不友好.

准确率如何保证,如何调整超参数,只有认真研究算法机理,才能真正清除内部实现机制后给予改进.

下面,上代码:

?
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
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
import os
 
import sys
import argparse
import json
import cv2
import numpy as np
from sklearn.cluster import KMeans
# from star_detector import StarFeatureDetector
from sklearn.ensemble import ExtraTreesClassifier
from sklearn import preprocessing
 
try :
  import cPickle as pickle #python 2
except ImportError as e:
  import pickle #python 3
 
def load_training_data(input_folder):
  training_data = []
  if not os.path.isdir(input_folder):
   raise IOError( "The folder " + input_folder + " doesn't exist" )
  
  for root, dirs, files in os.walk(input_folder):
   for filename in (x for x in files if x.endswith( '.jpg' )):
    filepath = os.path.join(root, filename)
    print (filepath)
    object_class = filepath.split( '\\' )[ - 2 ]
    print ( "object_class" ,object_class)
    training_data.append({ 'object_class' : object_class, 'image_path' : filepath})
     
  return training_data
class StarFeatureDetector( object ):
  def __init__( self ):
   self .detector = cv2.xfeatures2d.StarDetector_create()
  def detect( self , img):
   return self .detector.detect(img)
 
class FeatureBuilder( object ):
  def extract_features( self , img):
   keypoints = StarFeatureDetector().detect(img)
   keypoints, feature_vectors = compute_sift_features(img, keypoints)
   return feature_vectors
  def get_codewords( self , input_map, scaling_size, max_samples = 12 ):
   keypoints_all = []
   count = 0
   cur_class = ''
   for item in input_map:
    if count > = max_samples:
     if cur_class ! = item[ 'object_class' ]:
      count = 0
     else :
      continue
    count + = 1
    if count = = max_samples:
     print ( "Built centroids for" , item[ 'object_class' ])
 
    cur_class = item[ 'object_class' ]
    img = cv2.imread(item[ 'image_path' ])
    img = resize_image(img, scaling_size)
    num_dims = 128
    feature_vectors = self .extract_features(img)
    keypoints_all.extend(feature_vectors)
 
   kmeans, centroids = BagOfWords().cluster(keypoints_all)
   return kmeans, centroids
class BagOfWords( object ):
  def __init__( self , num_clusters = 32 ):
   self .num_dims = 128
   self .num_clusters = num_clusters
   self .num_retries = 10
 
  def cluster( self , datapoints):
   kmeans = KMeans( self .num_clusters,
       n_init = max ( self .num_retries, 1 ),
       max_iter = 10 , tol = 1.0 )
   res = kmeans.fit(datapoints)
   centroids = res.cluster_centers_
   return kmeans, centroids
 
  def normalize( self , input_data):
   sum_input = np. sum (input_data)
 
   if sum_input > 0 :
    return input_data / sum_input
   else :
    return input_data
  def construct_feature( self , img, kmeans, centroids):
   keypoints = StarFeatureDetector().detect(img)
   keypoints, feature_vectors = compute_sift_features(img, keypoints)
   labels = kmeans.predict(feature_vectors)
   feature_vector = np.zeros( self .num_clusters)
 
   for i, item in enumerate (feature_vectors):
    feature_vector[labels[i]] + = 1
 
   feature_vector_img = np.reshape(feature_vector, (( 1 , feature_vector.shape[ 0 ])))
   return self .normalize(feature_vector_img)
# Extract features from the input images and
# map them to the corresponding object classes
def get_feature_map(input_map, kmeans, centroids, scaling_size):
  feature_map = []
  for item in input_map:
   temp_dict = {}
   temp_dict[ 'object_class' ] = item[ 'object_class' ]
 
   print ( "Extracting features for" , item[ 'image_path' ])
   img = cv2.imread(item[ 'image_path' ])
   img = resize_image(img, scaling_size)
 
   temp_dict[ 'feature_vector' ] = BagOfWords().construct_feature(img, kmeans, centroids)
   if temp_dict[ 'feature_vector' ] is not None :
    feature_map.append(temp_dict)
  return feature_map
 
# Extract SIFT features
def compute_sift_features(img, keypoints):
  if img is None :
   raise TypeError( 'Invalid input image' )
 
  img_gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
  keypoints, descriptors = cv2.xfeatures2d.SIFT_create().compute(img_gray, keypoints)
  return keypoints, descriptors
 
# Resize the shorter dimension to 'new_size'
# while maintaining the aspect ratio
def resize_image(input_img, new_size):
  h, w = input_img.shape[: 2 ]
  scaling_factor = new_size / float (h)
 
  if w < h:
   scaling_factor = new_size / float (w)
 
  new_shape = ( int (w * scaling_factor), int (h * scaling_factor))
  return cv2.resize(input_img, new_shape)
 
def build_features_main():
  data_folder = 'training_images\\'
  scaling_size = 200
  codebook_file = 'codebook.pkl'
  feature_map_file = 'feature_map.pkl'
  # Load the training data
  training_data = load_training_data(data_folder)
 
  # Build the visual codebook
  print ( "====== Building visual codebook ======" )
  kmeans, centroids = FeatureBuilder().get_codewords(training_data, scaling_size)
  if codebook_file:
   with open (codebook_file, 'wb' ) as f:
    pickle.dump((kmeans, centroids), f)
 
  # Extract features from input images
  print ( "\n====== Building the feature map ======" )
  feature_map = get_feature_map(training_data, kmeans, centroids, scaling_size)
  if feature_map_file:
   with open (feature_map_file, 'wb' ) as f:
    pickle.dump(feature_map, f)
# --feature-map-file feature_map.pkl --model- file erf.pkl
#----------------------------------------------------------------------------------------------------------
class ERFTrainer( object ):
  def __init__( self , X, label_words):
   self .le = preprocessing.LabelEncoder()
   self .clf = ExtraTreesClassifier(n_estimators = 100 ,
     max_depth = 16 , random_state = 0 )
 
   y = self .encode_labels(label_words)
   self .clf.fit(np.asarray(X), y)
 
  def encode_labels( self , label_words):
   self .le.fit(label_words)
   return np.array( self .le.transform(label_words), dtype = np.float32)
 
  def classify( self , X):
   label_nums = self .clf.predict(np.asarray(X))
   label_words = self .le.inverse_transform([ int (x) for x in label_nums])
   return label_words
#------------------------------------------------------------------------------------------
 
class ImageTagExtractor( object ):
  def __init__( self , model_file, codebook_file):
   with open (model_file, 'rb' ) as f:
    self .erf = pickle.load(f)
 
   with open (codebook_file, 'rb' ) as f:
    self .kmeans, self .centroids = pickle.load(f)
 
  def predict( self , img, scaling_size):
   img = resize_image(img, scaling_size)
   feature_vector = BagOfWords().construct_feature(
     img, self .kmeans, self .centroids)
   image_tag = self .erf.classify(feature_vector)[ 0 ]
   return image_tag
 
def train_Recognizer_main():
  feature_map_file = 'feature_map.pkl'
  model_file = 'erf.pkl'
  # Load the feature map
  with open (feature_map_file, 'rb' ) as f:
   feature_map = pickle.load(f)
  # Extract feature vectors and the labels
  label_words = [x[ 'object_class' ] for x in feature_map]
  dim_size = feature_map[ 0 ][ 'feature_vector' ].shape[ 1 ]
  X = [np.reshape(x[ 'feature_vector' ], (dim_size,)) for x in feature_map]
 
  # Train the Extremely Random Forests classifier
  erf = ERFTrainer(X, label_words)
  if model_file:
   with open (model_file, 'wb' ) as f:
    pickle.dump(erf, f)
  #--------------------------------------------------------------------
  # args = build_arg_parser().parse_args()
  model_file = 'erf.pkl'
  codebook_file = 'codebook.pkl'
  import os
  rootdir = r "F:\airplanes"
  list = os.listdir(rootdir)
  for i in range ( 0 , len ( list )):
   path = os.path.join(rootdir, list [i])
   if os.path.isfile(path):
    try :
     print (path)
     input_image = cv2.imread(path)
     scaling_size = 200
     print ( "\nOutput:" , ImageTagExtractor(model_file,codebook_file)\
       .predict(input_image, scaling_size))
    except :
     continue
  #-----------------------------------------------------------------------
build_features_main()
train_Recognizer_main()

以上这篇Python构建图像分类识别器的方法就是小编分享给大家的全部内容了,希望能给大家一个参考,也希望大家多多支持我.

原文链接:https://blog.csdn.net/weixin_42039090/article/details/80673711 。

最后此篇关于Python构建图像分类识别器的方法的文章就讲到这里了,如果你想了解更多关于Python构建图像分类识别器的方法的内容请搜索CFSDN的文章或继续浏览相关文章,希望大家以后支持我的博客! 。

24 4 0
Copyright 2021 - 2024 cfsdn All Rights Reserved 蜀ICP备2022000587号
广告合作:1813099741@qq.com 6ren.com