def loadDataSet():
if not os.path.exists('tmp.pkl'):
samples = []
# 遍历数据集目录
for img_name in sorted(glob.glob('datasets/original/face/*.jpg')):
# 读取正样本图像
print(img_name)
img = np.array(Image.open(img_name).resize((24, 24)).convert("L"))
# 提取NPD特征
features = NPDFeature(img).extract()
sample = np.r_[features, 1] # 在正样本特征后面加一个Label为1
samples.append(sample)
for img_name in sorted(glob.glob('datasets/original/nonface/*.jpg')):
# 读取负样本图像
print(img_name)
img = np.array(Image.open(img_name).resize((24, 24)).convert("L"))
# 提取NPD特征
features = NPDFeature(img).extract()
sample = np.r_[features, -1] # 在负样本特征后面加一个Label为-1
samples.append(sample)
# 数据集打乱
random.shuffle(samples)
dataset = np.array(samples)
with open('tmp.pkl', 'wb') as output:
pickle.dump(dataset, output, True)
with open('tmp.pkl', 'rb') as input:
dataset = pickle.load(input)
print(dataset.shape)
# 将数据集切分为训练集和验证集
X_train = dataset[:dataset.shape[0] * 3 // 4, :dataset.shape[1] - 1]
y_train = dataset[:dataset.shape[0] * 3 // 4, dataset.shape[1] - 1]
X_validation = dataset[dataset.shape[0] * 3 // 4:, :dataset.shape[1] - 1]
y_validation = dataset[dataset.shape[0] * 3 // 4:, dataset.shape[1] - 1]
return X_train, X_validation, y_train, y_validation
def getFeature(Path, savePath):
for file in os.listdir(Path):
file_abs_path = os.path.join(Path, file)
if os.path.isdir(file_abs_path):
getFeature(file_abs_path, savePath)
if os.path.isfile(file_abs_path):
if os.path.exists(file_abs_path):
#using NPDFeature class to get feature
im = np.array(Image.open(file_abs_path))
#print("image.shape = ", im.shape)
#print("**8*************======", im)
npdf = NPDFeature(im)
pic_features = npdf.extract()
print("pic_features", pic_features.shape)
#to get path for saving features as files
parent_path_name = os.path.dirname(file_abs_path).split("/")[-1]
save_path = os.path.join(savePath, parent_path_name)
save_path = os.path.join(save_path, os.path.splitext(file)[0])
if not os.path.exists(save_path):
#dump to file
output = open(save_path, "wb")
PROROCOL = 0
pickle.dump(pic_features, output, PROROCOL)
def extract_to_list(path_from):
result = []
for f in os.listdir(path_from):
i = Image.open(path_from + "/" + f)
im_array = np.array(i)
npd = NPDFeature(im_array)
features = npd.extract()
result.append(features)
return result
def get_npdArray_from_diskImg(pathDir):
list = []
file_names = os.listdir(pathDir)
for file_name in file_names:
img = Image.open('%s%s' % (pathDir, file_name))
img = img.resize((24, 24))
img = np.array(img.convert('L'))
npdFeature = NPDFeature(img)
npdArray = npdFeature.extract()
list.append(npdArray)
faces_npdArray = np.array(list)
return faces_npdArray
def exact_nonface():
#处理nonface
i=0;
feature=[]
label=[]
while i<500:
img_wait_deal=np.array(Image.open("/home/kodgv/第三次实验/ML2017-lab-03/huidutu/non" + str(i) + ".jpg"))
NPD=NPDFeature(img_wait_deal)
label.append(-1)
feature.append(list(NPD.extract()))
i+=1
np.save("nonfeature.npy",np.array(feature))
np.save("nonlabel.npy",np.array(label).reshape(1,len(label)))
def get_feature(path):
features = numpy.array([])
files = os.listdir(path)
for k in range(len(files)):
im = Image.open(path + files[k])
image = numpy.ones(shape=(24, 24), dtype=int)
for i in range(24):
for j in range(24):
image[i][j] = im.getpixel((i, j))
NPDFeature1 = NPDFeature(image)
feature = NPDFeature1.extract()
features = numpy.concatenate((features, feature))
return features
def npd_features(imgs):
features = np.empty(shape=(0,
SCALE_SIZE[0] * (SCALE_SIZE[0] - 1) // 2))
for i in range(len(imgs)):
feature = NPDFeature((imgs[i] * 255).astype(np.int8)).extract()
features = np.vstack((features, feature))
print(features.shape)
return features
def extract_fea(img_dirs, img_labels, store_name):
'''
预处理阶段，处理为24*24，灰度图
'''
fea_list = []
for i in range(len(img_dirs)):
temp_img = io.imread(img_dirs[i])
temp_gray_img = color.rgb2gray(temp_img)
temp_resized_img = transform.resize(temp_gray_img, (24, 24))
temp_resized_img = img_as_ubyte(temp_resized_img)
#提取特征
npd_fea = NPDFeature(temp_resized_img)
temp_fea = npd_fea.extract()
temp_label = img_labels[i]
fea_list.append((temp_fea, temp_label))
o_file = open(store_name, 'wb')
pickle.dump(fea_list, o_file, -1)
o_file.close()
def getNPDFeature(imageList):
'''
函数作用：提取灰度格式图片的NPD特征
:param imageList:列表，元素是灰度格式图片
:return:列表，元素是每个图片的NPD特征
'''
#temp = NPDFeature(imageList)
featureList = [NPDFeature(im).extract() for im in imageList]
return featureList
def get_grayscale(path, x=None, y=None):
for file in os.listdir(path):
file_path = os.path.join(path, file)
labels = None
if path == 'datasets/original/face/':
labels = [1]
elif path == 'datasets/original/nonface/':
labels = [-1]
image = Image.open(file_path).convert('L').resize(
(24, 24)) #将rgb图片转化成灰度图
if (x is None) & (y is None):
x = np.array([NPDFeature(np.asarray(image)).extract()])
y = np.array([labels])
else:
#把脸的和非脸的图片拼接起来
x = np.vstack((x, NPDFeature(np.asarray(image)).extract()))
y = np.vstack((y, labels))
return x, y
def to_gray_resize(path, x=None, y=None):
for file in os.listdir(path):
file_path = os.path.join(path, file)
label = None
if path == 'datasets/original/face/':
label = [1]
elif path == 'datasets/original/nonface/':
label = [-1]
image = Image.open(file_path).convert('L').resize(
(24, 24)) #打开图片并转换成灰度图
if (x is None) & (y is None): #不能使用==：==是element-wise的
x = np.array([NPDFeature(np.asarray(image)).extract()])
y = np.array([label])
else:
#把是脸的和非脸的图片都拼接起来
x = np.vstack((x, NPDFeature(
np.asarray(image)).extract())) #与np.concat([x1, x2], axis=0)等价
y = np.vstack((y, label))
return x, y
def convert_images(path, label):
imgs = os.listdir(path)
X = []
y = []
for img in imgs:
img_path = path + "/" + img
im = Image.open(img_path).convert("L").resize((24, 24))
X.append(np.array(NPDFeature(np.asarray(im)).extract()))
y.append(label)
X = np.array(X).reshape(len(imgs), -1)
y = np.array(y).reshape(len(imgs), 1)
return X, y
def pre_process(dir, file):
features = np.array([])
for filename in os.listdir(dir):
img = Image.open(os.path.join(dir, filename))
resize_img = img.resize((24, 24))
gray_img = np.array(resize_img.convert("L"))
feature = NPDFeature(gray_img).extract()
features = np.append(features, feature).reshape(-1, 165600)
print(features.shape)
pass
with open(file, "wb") as f:
pickle.dump(features, f)
def get_data():
List = []
facelist = os.listdir('./datasets/original/face/') #
for i in range(len(facelist)):
location = './datasets/original/face/' + facelist[i]
im = Image.open(location).convert('L') #读取图片，将全部图片转成大小为24*24的灰度图
array = np.array(im).astype(float)
array = imresize(
array, (24, 24)) # Convert between PIL image and NumPy ndarray
#im2 = Image.fromarray(array) # Convert between PIL image and NumPy ndarray
npdfeature = NPDFeature(array) #处理数据集数据，提取NPD特征。
feature = npdfeature.extract()
List.append(feature)
nonfacelist = os.listdir('./datasets/original/nonface/') #
for i in range(len(nonfacelist)):
location = './datasets/original/nonface/' + nonfacelist[i]
im = Image.open(location).convert('L') #读取图片，将全部图片转成大小为24*24的灰度图
array = np.array(im).astype(float)
array = imresize(
array, (24, 24)) # Convert between PIL image and NumPy ndarray
#im2 = Image.fromarray(array) # Convert between PIL image and NumPy ndarray
npdfeature = NPDFeature(array) #处理数据集数据，提取NPD特征。
feature = npdfeature.extract()
List.append(feature)
file = open('feature.txt', 'wb') # 用pickle库中的dump()函数将预处理后的特征数据保存到缓存中
pickle.dump(List, file)
file.close()
def mk_dataset():
face_dir = "/Users/zoushuai/Python/lab/datasets/original/face"
nonface_dir = "/Users/zoushuai/Python/lab/datasets/original/nonface"
face_list = resize_image(face_dir)
nonface_list = resize_image(nonface_dir)
train_set = face_list[0:250]
train_set.extend(
nonface_list[0:250]) # trainset contains 250 faces and 250 nonfaces
train_set = np.array(train_set)
validate_set = face_list[250:500]
validate_set.extend(nonface_list[250:500]
) # validateset contains 250 faces and 250 nonfaces
validate_set = np.array(validate_set)
train_img2feature_list = []
validate_img2feature_list = []
for i in range(500):
npdFeature_train = NPDFeature(train_set[i])
train_img2feature_list.append(npdFeature_train.extract())
npdFeature_validate = NPDFeature(validate_set[i])
validate_img2feature_list.append(npdFeature_validate.extract())
train_img2feature_list = np.array(train_img2feature_list)
validate_img2feature_list = np.array(validate_img2feature_list)
AdaBoostClassifier.save(train_img2feature_list, 'train')
AdaBoostClassifier.save(validate_img2feature_list, 'validate')
def preprocess_data(dataset):
'''convert image to greyscale,resize to (24,24) and extract NPD feature'''
# feature size: 165600
num_feature = 165600
num_sample = dataset.shape[0]
dataset_processed = np.ndarray(shape=(num_sample, num_feature),
dtype=np.float32)
for i in range(num_sample):
img_grey = convert_to_grey(dataset[i])
img_grey_resize = imresize(img_grey, (24,24))
npdfeature = NPDFeature(img_grey_resize).extract()
dataset_processed[i] = npdfeature
return dataset_processed
def extract_feature(img_path, feature_path):
assert os.path.exists(img_path), "image path is not exist"
sum = 0 # 特征总数
fds = [] # 特征列表
for childDir in os.listdir(img_path):
f = os.path.join(img_path, childDir)
pil_im = Image.open(f).convert('L').resize((24, 24), Image.ANTIALIAS)
im = np.array(pil_im)
im_feat = NPDFeature(im).extract()
fds.append(im_feat)
sum += 1
joblib.dump(fds, feature_path)
print("%d sample features are extracted and saved." % sum)
return fds
def read_data():
X = []
y = []
for i in range(500):
path = get_i_face_image_path(i)
image = Image.open(path)
image = image.convert('L')
image = image.resize((16, 16))
X.append(image)
y.append(1)
for i in range(500):
path = get_i_nonface_image_path(i)
image = Image.open(path)
image = image.convert('L')
image = image.resize((16, 16))
X.append(image)
y.append(-1)
feature = []
for i in tqdm(range(len(X)), desc='pre_train', leave=True):
array_image = np.array(X[i])
fea = NPDFeature(array_image)
feature.append(fea.extract())
return feature, y
def load_data():
face_path = u'C:/Users/47864/Desktop/Data/datasets/original/face'
nonface_path = u'C:/Users/47864/Desktop/Data/datasets/original/nonface'
face_image = os.listdir(face_path)
nonface_image = os.listdir(nonface_path)
num_face_image = len(face_image)
num_nonface_image = len(nonface_image)
dataset = []
for i in range(num_face_image):
img = Image.open(face_path + '/' + face_image[i])
img = img.convert('L')
img = img.resize((24, 24), Image.ANTIALIAS)
img = NPDFeature(np.array(img))
dataset.append(np.concatenate((img.extract(), np.array([1]))))
for i in range(num_nonface_image):
img = Image.open(nonface_path + '/' + nonface_image[i])
img = img.convert('L')
img = img.resize((24, 24), Image.ANTIALIAS)
img = NPDFeature(np.array(img))
dataset.append(np.concatenate((img.extract(), np.array([-1]))))
return dataset
def readimg():
currentpath1='./datasets/original/face/face_'
currentpath2='./datasets/original/nonface/nonface_'
for i in range(0,500):
img_face = mpimg.imread(currentpath1+"{:0>3d}".format(i)+".jpg")
img_face_=rgb2gray(img_face)
f=NPDFeature(img_face_)
feature_=f.extract()
feature.append(feature_)
label.append(1)
for i in range(0,500):
img_nonface=mpimg.imread(currentpath2+"{:0>3d}".format(i)+".jpg")
img_nonface_=rgb2gray(img_nonface)
f=NPDFeature(img_nonface_)
feature_=f.extract()
feature.append(feature_)
label.append(-1)
def preprocess_image():
os.makedirs("datasets\\features")
# preprocess face images
for i in range(500):
# set the fetch address
fetch_address = "datasets\\original\\face\\face_%03d.jpg" % i
# read jpg file
# im = matplotlib.image.imread(fetch_address)
# transform to the 24*24 gray image
# xi = transform_image(im)
# create a NPDFeature class
xi = Image.open(fetch_address).convert('L').resize((24, 24))
xi = np.array(xi)
npd = NPDFeature(xi)
# extract feature
feature = npd.extract()
# set the filename
filename = "datasets\\features\\face%03d.pickle" % i
# save the feature in pickle file
save_data(filename, feature)
# preprocess nonface images
for i in range(500):
# set the fetch address
fetch_address = "datasets\\original\\nonface\\nonface_%03d.jpg" % i
# read jpg file
# im = matplotlib.image.imread(fetch_address)
# transform to the 24*24 gray image
# xi = transform_image(im)
# create a NPDFeature class
xi = Image.open(fetch_address).convert('L').resize((24, 24))
xi = np.array(xi)
npd = NPDFeature(xi)
# extract feature
feature = npd.extract()
# set the filename
filename = "datasets\\features\\nonface%03d.pickle" % i
# save the feature in pickle file
save_data(filename, feature)
def Feature_extract():
#提取特征
face_path = '.\\datasets\\original\\face\\face_%03d.jpg'
faces_path = []
for i in range(500):
faces_path.append(face_path % i)
nonface_path = '.\\datasets\\original\\nonface\\nonface_%03d.jpg'
nonfaces_path = []
for i in range(500):
nonfaces_path.append(nonface_path % i)
train = np.zeros((1000, 165600))
for i in range(500):
img = Image.open(faces_path[i])
img = img.convert('L').resize((24, 24))
nf = NPDFeature(np.array(img))
train[i * 2] = nf.extract()
img = Image.open(nonfaces_path[i])
img = img.convert('L').resize((24, 24))
nf = NPDFeature(np.array(img))
train[i * 2 + 1] = nf.extract()
AdaBoostClassifier.save(train, 'train.txt')
def getFeature(image):
NPD = NPDFeature(image)
return NPD
nonfacepath = 'datasets/original/nonface'
face = []
nonface = []
#for each image, convert it into grayscale presentation
#scale to 24x24
#and extract its NPD feature
facedir = os.listdir(facepath)
for i in range(0, len(facedir)):
if facedir[i].endswith('jpg'):
path = os.path.join(facepath, facedir[i])
img = mpimg.imread(path)
img = rgb2gray(img)
img = misc.imresize(img, [24, 24])
face.append(NPDFeature(img).extract())
nonfacedir = os.listdir(nonfacepath)
for i in range(0, len(nonfacedir)):
if nonfacedir[i].endswith('jpg'):
path = os.path.join(nonfacepath, nonfacedir[i])
img = mpimg.imread(path)
img = rgb2gray(img)
img = misc.imresize(img, [24, 24])
nonface.append(NPDFeature(img).extract())
X = np.array(face + nonface)
y = np.ones([1000])
y[500:999] = -1
X_train, X_vali, y_train, y_vali = train_test_split(X,
def extra_img_features():
for i in range(0, len(img)):
f = NPDFeature(img[i])
features = f.extract()
img_features.append(features)
n_face = 0
n_noface = 0
for file in os.listdir(pos_img):
n_face+=1
for file in os.listdir(neg_img):
n_noface+=1
pos_ds = []
n=0
for file in os.listdir(pos_img):
n+=1
print("Face: %4d / %4d" %(n, n_face),end='')
sys.stdout.write('\r')
img = cv2.imread(os.path.join(pos_img,file),cv2.IMREAD_GRAYSCALE) # load image file in Gray_mode
img = cv2.resize(img,(24,24))
feature = NPDFeature(img).extract()
pos_ds.append(feature)
pos_ds = np.asarray(pos_ds)
print(pos_ds.shape)
neg_ds = []
n=0
for file in os.listdir(neg_img):
n += 1
print("No Face: %4d / %4d" %(n, n_noface),end='')
sys.stdout.write('\r')
img = cv2.imread(os.path.join(neg_img,file),cv2.IMREAD_GRAYSCALE)
img = cv2.resize(img,(24,24))
feature = NPDFeature(img).extract()
neg_ds.append(feature)
neg_ds = np.asarray(neg_ds)
from PIL import Image
import pylab
import numpy as np
import os
from feature import NPDFeature
import pickle
def save(model, filename):
with open(filename, "wb") as f:
pickle.dump(model, f)
pos_img_path = './datasets/original/nonface/'
#pos_img_path = '.\\datasets\\original\\nonface'
#pos_img_path = '/Users/limuyi/Downloads/study/大三上/机器学习/6班谭明奎全英班/机器学习实验/机器学习实验内容/3AdaBoost人脸分类/ML2017-lab-03/datasets/original/nonface'
features_face = []
pos_list_dir = os.listdir(pos_img_path)
count = 1
for filename in pos_list_dir:
im = Image.open(pos_img_path + filename).convert('L').resize((24, 24))
im_array = np.array(im)
im_feature = NPDFeature(im_array).extract()
print('pos:', count)
count += 1
features_face.append(im_feature)
save(features_face, 'features_nonface')
def get_features(img):
img = resize_image(img)
features = NPDFeature(img).extract()
return features
from sklearn.model_selection import train_test_split
from ensemble import AdaBoostClassifier
from feature import NPDFeature
from sklearn.tree import DecisionTreeClassifier
from sklearn.metrics import classification_report
if __name__ == "__main__":
# write your code here
X = []
y = []
for i in range(0, 500):
path = "C:\\Users\\Administrator\\Desktop\\ML2017-lab-03\\datasets\\original\\face\\face_%.3d.jpg" % (
i)
img = Image.open(path).convert('L').resize((24, 24))
X.append(NPDFeature(np.array(img)).extract())
y.append(1)
print(i)
for i in range(0, 500):
path = "C:\\Users\\Administrator\\Desktop\\ML2017-lab-03\\datasets\\original\\nonface\\nonface_%.3d.jpg" % (
i)
img = Image.open(path).convert('L').resize((24, 24))
X.append(NPDFeature(np.array(img)).extract())
y.append(-1)
print(i)
X_train, X_test, y_train, y_test = train_test_split(X,
y,
test_size=0.33,
random_state=42)
def npd_feature():
for i in range(0, len(imgs)):
print(i)
features = NPDFeature(imgs[i]).extract()
img_features.append(features)
