使用SIFT特征提取和K-Means方法对图片进行分类

由于项目的需要，需要搜集一批有标签的图片，但是人力没有那么多，无法对图片进行分类，所以就先用无监督的方法对用机器对图片自动分类，先富集一批数据，然后再对模型进行训练，于是就想到了k-means算法，但是图片需要提取特征，于是想到了使用SIFT来对图片进行提取特征，提取的方法使用OpenCV的库来进行提取，具体安装OpenCV的方法请参考：点击打开链接。

废话不多说，看代码：

  
  
#-*- encoding:utf-8 -*-
__date__ = '17/04/21'
'''
CV_INTER_NN - 最近邻插值,
CV_INTER_LINEAR - 双线性插值 (缺省使用)
CV_INTER_AREA - 使用象素关系重采样。当图像缩小时候，该方法可以避免波纹出现。当图像放大时，类似于 CV_INTER_NN 方法..
CV_INTER_CUBIC - 立方插值
'''

import os, codecs
import cv2
import numpy as np
from sklearn.cluster import KMeans

def get_file_name(path):
'''
Args: path to list; Returns: path with filenames
'''
filenames = os.listdir(path)
path_filenames = []
filename_list = []
for file in filenames:
if not file.startswith( '.'):
path_filenames.append(os.path.join(path, file))
filename_list.append(file)

return path_filenames

def knn_detect(file_list, cluster_nums, randomState = None):
features = []
files = file_list
sift = cv2.SIFT()
for file in files:
print(file)
img = cv2.imread(file)
img = cv2.resize(img, ( 32, 32), interpolation=cv2.INTER_CUBIC)

gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
print(gray.dtype)
_, des = sift.detectAndCompute(gray, None)

if des is None:
file_list.remove(file)
continue

reshape_feature = des.reshape( -1, 1)
features.append(reshape_feature[ 0].tolist())

input_x = np.array(features)

kmeans = KMeans(n_clusters = cluster_nums, random_state = randomState).fit(input_x)

return kmeans.labels_, kmeans.cluster_centers_

def res_fit(filenames, labels):

files = [file.split( '/')[ -1] for file in filenames]

return dict(zip(files, labels))

def save(path, filename, data):
file = os.path.join(path, filename)
with codecs.open(file, 'w', encoding = 'utf-8') as fw:
for f, l in data.items():
fw.write( "{}\t{}\n".format(f, l))

def main():
path_filenames = get_file_name( "./picture/")

labels, cluster_centers = knn_detect(path_filenames, 2)

res_dict = res_fit(path_filenames, labels)
save( './', 'knn_res.txt', res_dict)

if __name__ == "__main__":
main()
  
  

使用的方法就是再path 里面传入picture的文件夹地址，还有需要分的类别数，然后程序检测过后将检测的结果写入文件。当然也可以根据检测结果将对应的图片写入对应的文件夹，这个就懒得弄了。还有就是可以设置初始化的rand_state。这个照着之前的维度设置就可以了，留作后期再弄。

-----------------------EOF--------------------------

参考文献：

http://scikit-learn.org/stable/modules/generated/sklearn.cluster.KMeans.html

http://opencv-python-tutroals.readthedocs.io/en/latest/py_tutorials/py_feature2d/py_sift_intro/py_sift_intro.html