def __init__(self, treshhold=20, single=False, logfile=None, pics_dir='data', isShow=True):
'''
single: 是否为处理单个图片的模式
logfile: 日志文件路径
pics_dir:若single为False则图片文件夹路径
若single为True则图片路径
isShow: 显示图片处理过程
'''
self.detector = Detector()
self.aligner = Aligner()
self.identifier = Identifier()
self.detector.set_min_face_size(treshhold)
if not single:
all_files = os.listdir(pics_dir)
for f in all_files:
if not os.path.isdir(f):
self.images_name.append(os.path.join(pics_dir,f))
else:
self.images_name.append(pics_dir)
self.images_name.sort()
if logfile == None:
logging.basicConfig(level=logging.INFO,
format='%(levelname)s-%(lineno)d-%(asctime)s %(message)s',
filename=logfile)
else:#print to screen
logging.basicConfig(level=logging.INFO,
format='%(levelname)s-%(lineno)d-%(asctime)s [FaceFeatureExtract]: %(message)s')
self.isShow = isShow
def test_cropface(model_path, img_name):
if model_path is None:
print('test_aligner: Please specify the path to the model folder')
return
if img_name is None:
print('test_aligner: Please specify the path to the image file')
return
assert os.path.isdir(
model_path
) is True, 'test_aligner: The model file path should be a folder'
assert os.path.isfile(img_name) is True, 'test_aligner: no such file'
model_path_fd = model_path + '/seeta_fd_frontal_v1.0.bin'
model_path_fa = model_path + '/seeta_fa_v1.1.bin'
model_path_fr = model_path + '/seeta_fr_v1.0.bin'
detector = Detector(model_path_fd)
detector.set_min_face_size(30)
aligner = Aligner(model_path_fa)
identifier = Identifier(model_path_fr)
image_color = Image.open(img_name).convert('RGB')
image_gray = image_color.convert('L')
import cv2
faces = detector.detect(image_gray)
for face in faces:
landmarks = aligner.align(image_gray, face)
crop_face = identifier.crop_face(image_color, landmarks)
Image.fromarray(crop_face).show()
identifier.release()
aligner.release()
detector.release()
def test_cropface():
detector = Detector()
detector.set_min_face_size(30)
aligner = Aligner()
identifier = Identifier()
image_color = Image.open('data/chloecalmon.png').convert('RGB')
image_gray = image_color.convert('L')
faces = detector.detect(image_gray)
for face in faces:
landmarks = aligner.align(image_gray, face)
crop_face = identifier.crop_face(image_color, landmarks)
Image.fromarray(crop_face).show()
identifier.release()
aligner.release()
detector.release()
def test_cropface():
detector = Detector('SeetaFaceEngine/model/seeta_fd_frontal_v1.0.bin')
detector.set_min_face_size(30)
aligner = Aligner('SeetaFaceEngine/model/seeta_fa_v1.1.bin')
identifier = Identifier('SeetaFaceEngine/model/seeta_fr_v1.0.bin')
image_color = Image.open('data/chloecalmon.png').convert('RGB')
image_gray = image_color.convert('L')
import cv2
faces = detector.detect(image_gray)
for face in faces:
landmarks = aligner.align(image_gray, face)
crop_face = identifier.crop_face(image_color, landmarks)
Image.fromarray(crop_face).show()
identifier.release()
aligner.release()
detector.release()
def test_recognition(img_path,db_path):
if not os.path.isfile(img_path) or not os.path.isdir(db_path):
print("indicated path doesn't exist!");return
# load model
detector = Detector(DET_MODEL_PATH)
aligner = Aligner(ALI_MODEL_PATH)
identifier = Identifier(REC_MODEL_PATH)
# load test image
image_color_A = cv2.imread(img_path, cv2.IMREAD_COLOR)
image_gray_A = cv2.cvtColor(image_color_A, cv2.COLOR_BGR2GRAY)
faces_A = detector.detect(image_gray_A)
# load database
for fn in os.listdir(db_path):
fp = os.path.join(db_path,fn)
if not os.path.isfile(fp): continue
image_color_B = cv2.imread(fp, cv2.IMREAD_COLOR)
image_gray_B = cv2.cvtColor(image_color_B, cv2.COLOR_BGR2GRAY)
# detect face in image
faces_B = detector.detect(image_gray_B)
if len(faces_A) and len(faces_B):
landmarks_A = aligner.align(image_gray_A, faces_A[0])
featA = identifier.extract_feature_with_crop(image_color_A, landmarks_A)
# cv2.rectangle(image_color_A, (faces_A[0].left, faces_A[0].top), (faces_A[0].right, faces_A[0].bottom), (0,255,0), thickness=2)
sim_list = []
for face in faces_B:
landmarks_B = aligner.align(image_gray_B, face)
featB = identifier.extract_feature_with_crop(image_color_B, landmarks_B)
sim = identifier.calc_similarity(featA, featB)
sim_list.append(sim)
print('sim: {}'.format(sim_list))
# index = np.argmax(sim_list)
for i, face in enumerate(faces_B):
color = (0,255,0) if sim_list[i] > 0.5 else (0,0,255)
cv2.rectangle(image_color_B, (face.left, face.top), (face.right, face.bottom), color, thickness=2)
# cv2.imshow('test', resize(image_color_A))
cv2.imshow('double', resize(image_color_B))
cv2.waitKey(0)
release(aligner,detector,identifier)
def test_identifier():
print('test identifier:')
detector = Detector()
aligner = Aligner()
identifier = Identifier()
# load image
image_color_A = Image.open('data/single.jpg').convert('RGB')
image_gray_A = image_color_A.convert('L')
image_color_B = Image.open('data/double.jpg').convert('RGB')
image_gray_B = image_color_B.convert('L')
# detect face in image
faces_A = detector.detect(image_gray_A)
faces_B = detector.detect(image_gray_B)
draw_A = ImageDraw.Draw(image_color_A)
draw_B = ImageDraw.Draw(image_color_B)
if len(faces_A) and len(faces_B):
landmarks_A = aligner.align(image_gray_A, faces_A[0])
featA = identifier.extract_feature_with_crop(image_color_A,
landmarks_A)
print(len(featA))
draw_A.rectangle([(faces_A[0].left, faces_A[0].top),
(faces_A[0].right, faces_A[0].bottom)],
outline='green')
sim_list = []
for face in faces_B:
landmarks_B = aligner.align(image_gray_B, face)
featB = identifier.extract_feature_with_crop(
image_color_B, landmarks_B)
sim = identifier.calc_similarity(featA, featB)
sim_list.append(sim)
print('sim: {}'.format(sim_list))
index = np.argmax(sim_list)
for i, face in enumerate(faces_B):
color = 'green' if i == index else 'red'
draw_B.rectangle([(face.left, face.top),
(face.right, face.bottom)],
outline=color)
image_color_A.show()
image_color_B.show()
identifier.release()
aligner.release()
detector.release()
def test_identifier():
print('test identifier:')
# load model
detector = Detector()
aligner = Aligner()
identifier = Identifier()
# load image
image_color_A = cv2.imread('data/single.jpg', cv2.IMREAD_COLOR)
image_gray_A = cv2.cvtColor(image_color_A, cv2.COLOR_BGR2GRAY)
image_color_B = cv2.imread('data/double.jpg', cv2.IMREAD_COLOR)
image_gray_B = cv2.cvtColor(image_color_B, cv2.COLOR_BGR2GRAY)
# detect face in image
faces_A = detector.detect(image_gray_A)
faces_B = detector.detect(image_gray_B)
if len(faces_A) and len(faces_B):
landmarks_A = aligner.align(image_gray_A, faces_A[0])
featA = identifier.extract_feature_with_crop(image_color_A,
landmarks_A)
cv2.rectangle(image_color_A, (faces_A[0].left, faces_A[0].top),
(faces_A[0].right, faces_A[0].bottom), (0, 255, 0),
thickness=2)
sim_list = []
for face in faces_B:
landmarks_B = aligner.align(image_gray_B, face)
featB = identifier.extract_feature_with_crop(
image_color_B, landmarks_B)
sim = identifier.calc_similarity(featA, featB)
sim_list.append(sim)
print('sim: {}'.format(sim_list))
index = np.argmax(sim_list)
for i, face in enumerate(faces_B):
color = (0, 255, 0) if i == index else (0, 0, 255)
cv2.rectangle(image_color_B, (face.left, face.top),
(face.right, face.bottom),
color,
thickness=2)
cv2.imshow('single', image_color_A)
cv2.imshow('double', image_color_B)
cv2.waitKey(0)
identifier.release()
aligner.release()
detector.release()
class SeetaFaceExtractor(BaseFeatureExtractor):
def __init__(self):
self.detector = Detector()
self.detector.set_min_face_size(30)
self.aligner = Aligner()
self.identifier = Identifier()
def extact(self, image_path, is_one_face=False):
image = cv2.imread(image_path)
if image is None:
print('Unable to load image: {}'.format(image_path))
return None
image_gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
if is_one_face:
bb1 = self.detector.detect(image_gray)
bbs = [bb1]
else:
bbs = self.detector.detect(image_gray)
if len(bbs) > 1:
print('More than two faces in {}'.format(image_path))
return None
if len(bbs) == 0 or (is_one_face and bb1 is None):
return None
reps = []
for bb in bbs:
aligned_face = self.aligner.align(image_gray, bb)
if aligned_face is None:
return None
rep = self.identifier.extract_feature_with_crop(image, aligned_face)
return rep
reps.append(rep)
# 因为已经明确一张图只有一个人
if is_one_face:
return reps[0]
else:
return reps
class FaceFeatureExtract():
'''
提取图片中人脸特征并存储
2018/3/30 V1.0
'''
images_name = []
isShow = None
detector = None
aligner = None
identifier = None
def __init__(self, treshhold=20, single=False, logfile=None, pics_dir='data', isShow=True):
'''
single: 是否为处理单个图片的模式
logfile: 日志文件路径
pics_dir:若single为False则图片文件夹路径
若single为True则图片路径
isShow: 显示图片处理过程
'''
self.detector = Detector()
self.aligner = Aligner()
self.identifier = Identifier()
self.detector.set_min_face_size(treshhold)
if not single:
all_files = os.listdir(pics_dir)
for f in all_files:
if not os.path.isdir(f):
self.images_name.append(os.path.join(pics_dir,f))
else:
self.images_name.append(pics_dir)
self.images_name.sort()
if logfile == None:
logging.basicConfig(level=logging.INFO,
format='%(levelname)s-%(lineno)d-%(asctime)s %(message)s',
filename=logfile)
else:#print to screen
logging.basicConfig(level=logging.INFO,
format='%(levelname)s-%(lineno)d-%(asctime)s [FaceFeatureExtract]: %(message)s')
self.isShow = isShow
def detect(self):
'''
返回一个字典数组：长度等于图片数。例如：
[
{
'name':'1.jpg'，
'features':[[1,23,...],[2,3,....]],
'landmarks':[[1,2,3,4],[5,6,7,8]]
},...
]
'''
pic_dic_list = []
for pic in self.images_name:
image = Image.open(pic).convert('RGB')
lands, faces, feats = self.__extract_features(image)
logging.info("detecting %s: find %d faces"%(pic, len(lands)))
if self.isShow:
draw = ImageDraw.Draw(image)
for i, face in enumerate(faces):
draw.rectangle([face.left, face.top, face.right, face.bottom], outline='red')
image.show()
pic_dic = {}
pic_dic['name'] = pic
pic_dic['landmarks'] = lands
pic_dic['feats'] = feats
pic_dic_list.append(pic_dic)
return pic_dic_list
def __extract_features(self, img):
# detect face in image
image_gray = img.convert('L')
faces = self.detector.detect(image_gray)
feats = []
landmarks = []
for detect_face in faces:
landmark = self.aligner.align(image_gray, detect_face)
landmarks.append(landmark)
feat = self.identifier.extract_feature_with_crop(img, landmark)
feats.append(feat)
return landmarks, faces, feats
def release(self):
'''
释放资源
'''
self.detector.release()
self.aligner.release()
self.identifier.release()
def recog(request):
try:
global avdata, name_img
detector = Detector('SeetaFaceEngine/model/seeta_fd_frontal_v1.0.bin')
aligner = Aligner('SeetaFaceEngine/model/seeta_fa_v1.1.bin')
identifier = Identifier('SeetaFaceEngine/model/seeta_fr_v1.0.bin')
detector.set_min_face_size(30)
if request.method == 'POST':
path = tempIMG(img=request.FILES['img'], )
path.save()
image_color_A = imread(str(path.img))
image_gray_A = cv2.cvtColor(image_color_A, cv2.COLOR_BGR2GRAY)
faces_A = detector.detect(image_gray_A)
cv2.rectangle(image_color_A, (faces_A[0].left, faces_A[0].top),
(faces_A[0].right, faces_A[0].bottom), (0, 255, 0),
thickness=2)
cv2.imwrite('facerecog/static/facerecog/' + 'img.jpg',
image_color_A)
length_list = []
if len(faces_A) or 0:
landmarks_A = aligner.align(image_gray_A, faces_A[0])
feat_test = identifier.extract_feature_with_crop(
image_color_A, landmarks_A)
average_sim_list = []
name_list = []
sim_list = []
for cla in avdata:
simlist = []
name_list.append(cla)
weight = 0
for fea in avdata[cla]:
sim = feat_match(feat_test, fea)
simlist.append(sim)
if sim > 0.5:
simlist.append(sim)
if sim > 0.55:
simlist.append(sim)
if sim > 0.6:
simlist.append(sim)
sim_list.append(simlist)
if len(simlist) == 0:
average_sim_list.append(0)
else:
average_sim = sum(simlist) / len(simlist)
average_sim_list.append(average_sim)
# print(average_sim_list)
max_index = average_sim_list.index(max(average_sim_list))
sort_list = sorted(average_sim_list)
result_list = []
for j in range(5):
result_list.append(name_list[average_sim_list.index(
sort_list[-(j + 1)])])
print(name_list[max_index])
print(average_sim_list[max_index])
identifier.release()
aligner.release()
detector.release()
name = str(request.FILES['img'])
print(name)
file_name = []
file_name.append(name)
print(result_list)
name_img = np.load('name_img.npy').item()
print(name_img[result_list[0]])
img_link = []
for name in result_list:
img_link.append(name_img[name])
content = {
'result_list': result_list,
'file_name': file_name,
'img_link': img_link
}
# return HttpResponse(json.dumps(content,ensure_ascii=False))
return render(request, 'facerecog/match.html', content)
except:
return HttpResponse('請指定有人臉的檔案!')
def __init__(self):
self.detector = Detector()
self.detector.set_min_face_size(30)
self.aligner = Aligner()
self.identifier = Identifier()
def test_identifier(model_path, img1_name, img2_name):
print('test identifier:')
if model_path is None:
print('test_identifier: Please specify the path to the model folder')
return
if img1_name is None:
print('test_identifier: Please specify the path to the image file1')
return
if img2_name is None:
print('test_identifier: Please specify the path to the image file2')
return
assert os.path.isdir(
model_path
) is True, 'test_identifier: The model file path should be a folder'
assert os.path.isfile(
img1_name) is True, 'test_identifier: no such image file1'
assert os.path.isfile(
img2_name) is True, 'test_identifier: no such image file2'
# load model
model_path_fd = model_path + '/seeta_fd_frontal_v1.0.bin'
model_path_fa = model_path + '/seeta_fa_v1.1.bin'
model_path_fr = model_path + '/seeta_fr_v1.0.bin'
detector = Detector(model_path_fd)
aligner = Aligner(model_path_fa)
identifier = Identifier(model_path_fr)
# load image
image_color_A = cv2.imread(img1_name, cv2.IMREAD_COLOR)
image_gray_A = cv2.cvtColor(image_color_A, cv2.COLOR_BGR2GRAY)
image_color_B = cv2.imread(img2_name, cv2.IMREAD_COLOR)
image_gray_B = cv2.cvtColor(image_color_B, cv2.COLOR_BGR2GRAY)
# detect face in image
faces_A = detector.detect(image_gray_A)
faces_B = detector.detect(image_gray_B)
if len(faces_A) and len(faces_B):
landmarks_A = aligner.align(image_gray_A, faces_A[0])
featA = identifier.extract_feature_with_crop(image_color_A,
landmarks_A)
cv2.rectangle(image_color_A, (faces_A[0].left, faces_A[0].top),
(faces_A[0].right, faces_A[0].bottom), (0, 255, 0),
thickness=2)
sim_list = []
for face in faces_B:
landmarks_B = aligner.align(image_gray_B, face)
featB = identifier.extract_feature_with_crop(
image_color_B, landmarks_B)
sim = identifier.calc_similarity(featA, featB)
sim_list.append(sim)
print('sim: {}'.format(sim_list))
index = np.argmax(sim_list)
for i, face in enumerate(faces_B):
color = (0, 255, 0) if i == index else (0, 0, 255)
cv2.rectangle(image_color_B, (face.left, face.top),
(face.right, face.bottom),
color,
thickness=2)
cv2.imshow('single', image_color_A)
cv2.imshow('double', image_color_B)
cv2.waitKey(0)
identifier.release()
aligner.release()
detector.release()
def test_identifier(model_path, img1_name, img2_name):
print('test identifier:')
if model_path is None:
print('test_identifier: Please specify the path to the model folder')
return
if img1_name is None:
print('test_identifier: Please specify the path to the image file1')
return
if img2_name is None:
print('test_identifier: Please specify the path to the image file2')
return
assert os.path.isdir(
model_path
) is True, 'test_identifier: The model file path should be a folder'
assert os.path.isfile(
img1_name) is True, 'test_identifier: no such image file1'
assert os.path.isfile(
img2_name) is True, 'test_identifier: no such image file2'
model_path_fd = model_path + '/seeta_fd_frontal_v1.0.bin'
model_path_fa = model_path + '/seeta_fa_v1.1.bin'
model_path_fr = model_path + '/seeta_fr_v1.0.bin'
detector = Detector(model_path_fd)
detector.set_min_face_size(30)
aligner = Aligner(model_path_fa)
identifier = Identifier(model_path_fr)
# load image
image_color_A = Image.open(img1_name).convert('RGB')
image_gray_A = image_color_A.convert('L')
image_color_B = Image.open(img2_name).convert('RGB')
image_gray_B = image_color_B.convert('L')
# detect face in image
faces_A = detector.detect(image_gray_A)
faces_B = detector.detect(image_gray_B)
draw_A = ImageDraw.Draw(image_color_A)
draw_B = ImageDraw.Draw(image_color_B)
if len(faces_A) and len(faces_B):
landmarks_A = aligner.align(image_gray_A, faces_A[0])
featA = identifier.extract_feature_with_crop(image_color_A,
landmarks_A)
draw_A.rectangle([(faces_A[0].left, faces_A[0].top),
(faces_A[0].right, faces_A[0].bottom)],
outline='green')
sim_list = []
for face in faces_B:
landmarks_B = aligner.align(image_gray_B, face)
featB = identifier.extract_feature_with_crop(
image_color_B, landmarks_B)
sim = identifier.calc_similarity(featA, featB)
sim_list.append(sim)
print('sim: {}'.format(sim_list))
index = np.argmax(sim_list)
for i, face in enumerate(faces_B):
color = 'green' if i == index else 'red'
draw_B.rectangle([(face.left, face.top),
(face.right, face.bottom)],
outline=color)
image_color_A.show()
image_color_B.show()
identifier.release()
aligner.release()
detector.release()
app = Flask(__name__)#, static_folder='static', static_url_path='')
ALI_MODEL_PATH = "models/seeta_fa_v1.1.bin"
DET_MODEL_PATH = "models/seeta_fd_frontal_v1.0.bin"
REC_MODEL_PATH = "models/seeta_fr_v1.0.bin"
from pyseeta import Detector
from pyseeta import Aligner
from pyseeta import Identifier
UPLOAD_FOLDER = 'examples/uploads'
WEBFILE_FOLDER = 'webfiles'
app.detector = Detector(DET_MODEL_PATH)
app.aligner = Aligner(ALI_MODEL_PATH)
app.identifier = Identifier(REC_MODEL_PATH)
app.results = []
app.config['UPLOAD_FOLDER'] = UPLOAD_FOLDER
app.config['WEBFILE_FOLDER'] = WEBFILE_FOLDER
app.config["CACHE_TYPE"] = "null"
app.config['ALLOWED_EXTENSIONS'] = set(['txt', 'png', 'jpg', 'jpeg', 'gif'])
app.config['MAX_CONTENT_LENGTH'] = 3 * 1024 * 1024 # 1 Mb limit
app.config['1.img'] = app.config['UPLOAD_FOLDER'] + "/1.img"
app.config['2.img'] = app.config['UPLOAD_FOLDER'] + "/2.img"
app.image_fn = os.path.join(app.config['UPLOAD_FOLDER'], "image.jpg")
app.result_fn = os.path.join(app.config['UPLOAD_FOLDER'], "result.txt")
app.filename = ""
def allowed_file(filename):
return '.' in filename and \