使用词袋模型对图像进行分类

# -*- coding: utf-8 -*-
import argparse as ap
import cv2
import imutils
import numpy as np
import os
from sklearn.externals import joblib
from scipy.cluster.vq import *
from sklearn import preprocessing
import math
import sys
import numpy
from sklearn import metrics
from sklearn.feature_extraction.text import HashingVectorizer
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.naive_bayes import MultinomialNB
from sklearn.feature_extraction.text import  CountVectorizer,TfidfTransformer
from sklearn.neighbors import KNeighborsClassifier
from sklearn.svm import SVC
from sklearn.svm import LinearSVC
import codecs
from sklearn.feature_selection import SelectPercentile, f_classif
from sklearn.feature_selection import SelectKBest
from sklearn.feature_selection import chi2
from sklearn import cross_validation

'''
使用词袋模型对图像进行分类：
1、数据格式，文件夹名即类别，每个文件夹下是一类图像
2、提取sift特征，将所有图像的sift特征放在一起，进行聚类，聚出n个视觉词
3、计算每幅图像有哪些视觉词，统计出词频矩阵
4、根据矩阵计算idf，进而得到tfidf矩阵，并对其进行L2归一化（向量中每个元素除以向量的L2范数->x/平方和开根号）
5、使用一般分类模型对其进行分类，计算P,R,F
'''

def load_data(path):
    '''
    每个文件夹下是一种图片
    :param path:种类文件夹路径
    :return: 图片路径列表和标签列表
    '''
    categories = os.listdir(path)
    img_pathes = []
    labels = []
    for path, dirs, files in os.walk(path):
        img_pathes.extend([os.path.join(path, file) for file in files])
        print path
        if len(files)>0:
            labels.extend([path.split('/')[-1]] * len(files))
    #print len(img_pathes),img_pathes
    #print len(labels),labels
    return img_pathes,labels

def cal_bow(image_paths, numWords):
    '''
    使用bag of word方法提取图像特征
    :param image_paths:
    :return:
    '''
    #numWords = 100
    #关键点检测对象
    fea_det = cv2.FeatureDetector_create("SIFT")
    #特征提取对象
    des_ext = cv2.DescriptorExtractor_create("SIFT")

    # List where all the descriptors are stored
    des_list = []

    for i, image_path in enumerate(image_paths):
        im = cv2.imread(image_path)
        print "Extract SIFT of %s image, %d of %d images" %(image_paths[i], i, len(image_paths))
        kpts = fea_det.detect(im)
        #可能存在没有检测出特征点的情况
        print len(kpts)
        #des有k行m列，每行代表一个特征，m是固定的特征维数
        kpts, des = des_ext.compute(im, kpts)
        des_list.append((image_path, des))

    # Stack all the descriptors vertically in a numpy array
    #image_path为图片路径，descriptor为对应图片的特征
    #将所有特征纵向堆叠起来,每行当做一个特征词
    descriptors = des_list[0][1]
    for image_path, descriptor in des_list[1:]:
        #vstack对矩阵进行拼接，将所有的特征word拼接到一起
        #print descriptor.shape, descriptors.shape
        #if descriptor != None:
        descriptors = np.vstack((descriptors, descriptor))

    #对特征词使用k-menas算法进行聚类
    print "Start k-means: %d words, %d key points" %(numWords, descriptors.shape[0])
    #最后输出的结果其实是两维的,第一维是聚类中心,第二维是损失distortion
    voc, variance = kmeans(descriptors, numWords, iter=1)

    #初始化一个bag of word矩阵，每行表示一副图像，每列表示一个视觉词，下面统计每副图像中视觉词的个数
    im_features = np.zeros((len(image_paths), numWords), "float32")
    for i in xrange(len(image_paths)):
        #计算每副图片的所有特征向量和voc中每个特征word的距离，返回为匹配上的word
        descriptor = des_list[i][1]
        #if descriptor != None:
        #根据聚类中心将所有数据进行分类des_list[i][1]为数据, voc则是kmeans产生的聚类中心.
        #vq输出有两个:一是各个数据属于哪一类的label,二是distortion
        words, distance = vq(des_list[i][1],voc)
        for w in words:
            im_features[i][w] += 1

    # Perform Tf-Idf vectorization
    nbr_occurences = np.sum( (im_features > 0) * 1, axis = 0)
    idf = np.array(np.log((1.0*len(image_paths)+1) / (1.0*nbr_occurences + 1)), 'float32')

    #L2归一化
    im_features = im_features*idf
    im_features = preprocessing.normalize(im_features, norm='l2')
    return im_features

def train_clf2(train_data, train_tags):
    #clf = SVC(kernel = 'linear')#default with 'rbf'
    clf = LinearSVC(C=1100.0)#default with 'rbf'
    clf.fit(train_data,train_tags)
    return clf

def evaluate(actual, pred):
    m_precision = metrics.precision_score(actual, pred, average="macro")
    m_recall = metrics.recall_score(actual, pred, average="macro")
    print 'precision:{0:.3f}'.format(m_precision)
    print 'recall:{0:0.3f}'.format(m_recall)
    print 'f1-score:{0:.8f}'.format(metrics.f1_score(actual,pred, average="macro"));
'''
'''
#提取图片特征并保存 图像大小为20*20
path = 'D:/data/Caltech_101/101_part/'
img_pathes,labels = load_data(path)
im_features = cal_bow(img_pathes, numWords=500)
joblib.dump((im_features, labels), "bof.pkl", compress=3)

#训练并测试
im_features, labels = joblib.load("bof.pkl")

X_train, X_test, y_train, y_test = \
    cross_validation.train_test_split(im_features,labels, test_size=0.3, random_state=0)
clf = train_clf2(X_train, y_train)

pred = clf.predict(X_test)
print pred
print y_test
evaluate(y_test, pred)

图像数据下载