from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
from scipy import misc
import tensorflow as tf
import numpy as np
import os
import copy
import argparse
import facenet
import align.detect_face
from flask import Flask
from flask import request
import json
import threading
import uuid
import base64
import cv2
import dlib
import sqllite
from concurrent.futures import ThreadPoolExecutor
#图片对比
def main(image):
emb = getEmb(image)
return imageExist(emb)
#图片检测
def test(image, flag):
# cv2读取图像
img = cv2.imread(image)
# 取灰度
img_gray = cv2.cvtColor(img, cv2.COLOR_RGB2GRAY)
# 人脸数rects
rects = detector(img_gray, 0)
landmarks = np.matrix([[p.x, p.y] for p in predictor(img, rects[0]).parts()])
dic = dict()
for idx, point in enumerate(landmarks):
# 68点的坐标
pos = (point[0, 0], point[0, 1])
dic[idx] = pos
# 利用cv2.circle给每个特征点画一个圈，共68个
# cv2.circle(img, pos, 1, color=(0, 255, 0))
# # 利用cv2.putText输出1-68
# font = cv2.FONT_HERSHEY_SIMPLEX
# cv2.putText(img, str(idx + 1), pos, font, 0.8, (0, 0, 255), 1, cv2.LINE_AA)
# cv2.namedWindow("img", 2)
# cv2.imshow("img", img)
# cv2.waitKey(0)
print("38", dic[37][0], dic[37][1])
print("42", dic[41][0], dic[41][1])
print("39", dic[38][0], dic[38][1])
print("41", dic[40][0], dic[40][1])
right1 = cal_len(dic[37][0], dic[37][1], dic[41][0], dic[41][1])
right2 = cal_len(dic[38][0], dic[38][1], dic[40][0], dic[40][1])
right = (right1+right2)*3
print("44", dic[43][0], dic[43][1])
print("48", dic[47][0], dic[47][1])
print("45", dic[44][0], dic[44][1])
print("47", dic[46][0], dic[46][1])
left1 = cal_len(dic[43][0], dic[43][1], dic[47][0], dic[47][1])
left2 = cal_len(dic[44][0], dic[44][1], dic[46][0], dic[46][1])
left = (left1+left2)*3
print("62", dic[61][0], dic[61][1])
print("68", dic[67][0], dic[67][1])
print("63", dic[62][0], dic[62][1])
print("67", dic[66][0], dic[66][1])
print("64", dic[63][0], dic[63][1])
print("66", dic[65][0], dic[65][1])
mouth1 = cal_len(dic[61][0], dic[61][1], dic[67][0], dic[67][1])
mouth2 = cal_len(dic[62][0], dic[62][1], dic[66][0], dic[66][1])
mouth3 = cal_len(dic[63][0], dic[63][1], dic[65][0], dic[65][1])
mouth = mouth1+mouth2+mouth3
print("37", dic[36][0], dic[36][1])
print("46", dic[45][0], dic[45][1])
print("28", dic[27][0], dic[27][1])
right_eye = cal_len(dic[36][0], dic[36][1], dic[27][0], dic[27][1])
left_eye = cal_len(dic[45][0], dic[45][1], dic[27][0], dic[27][1])
eye_eye = cal_len(dic[45][0], dic[45][1], dic[36][0], dic[36][1])
print("49", dic[48][0], dic[48][1])
print("55", dic[54][0], dic[54][1])
mouth_w = cal_len(dic[48][0], dic[48][1], dic[54][0], dic[54][1])
print("28", dic[27][0], dic[27][1])
print("34", dic[33][0], dic[33][1])
head = cal_len(dic[27][0], dic[27][1], dic[33][0], dic[33][1])
print("right_eye", right_eye) #右眼到中间间距
print("left_eye", left_eye) #左眼到中间间距
print("eye_eye", eye_eye) #两眼间距
print("eye_len_right", right) #眨右眼
print("eye_len_left", left) #眨左眼
print("mouth_len", mouth) #张嘴
print("mouth_w", mouth_w) #微笑
print("head", head) #点头
lens = {
"right_eye": right_eye,
"left_eye": left_eye,
"eye_eye": eye_eye,
"eye_len_right": right,
"eye_len_left": left,
"mouth_len": mouth,
"mouth_w": mouth_w,
"head": head
}
return lens
# #拍照检测
# def cutTest(image):
# # cv2读取图像
# img = cv2.imread(image)
# # 取灰度
# img_gray = cv2.cvtColor(img, cv2.COLOR_RGB2GRAY)
# # 人脸数rects
# rects = detector(img_gray, 0)
# landmarks = np.matrix([[p.x, p.y] for p in predictor(img, rects[0]).parts()])
# dic = dict()
# for idx, point in enumerate(landmarks):
# # 68点的坐标
# pos = (point[0, 0], point[0, 1])
# dic[idx] = pos
#
# eye_max = cal_len(dic[36][0], dic[36][1], dic[45][0], dic[45][1])
# eye_min = cal_len(dic[39][0], dic[39][1], dic[42][0], dic[42][1])
#
# param1 = eye_max/eye_min
#
# eye_nose = cal_len((dic[39][0]+dic[42][0])/2, (dic[39][1]+dic[42][1])/2,dic[33][0], dic[33][1])
# nose_low = cal_len(dic[33][0], dic[33][1], dic[8][0], dic[8][1])
#
# param2 = eye_nose/nose_low
#
# side1 = cal_len(dic[0][0], dic[0][1], dic[16][0], dic[16][1])
# side2 = cal_len(dic[2][0], dic[2][1], dic[14][0], dic[14][1])
# side3 = cal_len(dic[4][0], dic[4][1], dic[12][0], dic[12][1])
#
# param3 = side1/side2
# param4 = side2/side3
#
# nose = cal_len(dic[31][0], dic[31][1], dic[35][0], dic[35][1])
# mouth = cal_len(dic[48][0], dic[48][1], dic[54][0], dic[54][1])
#
# param5 = mouth/nose
#
# param6 = eye_max/mouth
# param7 = side2/mouth
#
# nose_mouth = cal_len(dic[33][0], dic[33][1], (dic[48][0]+dic[54][0])/2, (dic[48][1]+dic[54][1])/2)
# param8 = nose_low/nose_mouth
#
# a = cal_dif(500, param1, param2, param3, param4, param5, param6, param7, param8)
# param = round(np.log10(a), 3)
# return param
#拍照检测
def cutTest(image):
param = '-1'
emb = getEmb(image)
if emb != '-1':
param = imageExist(emb)
if param != '-1':
return param
imagels = str(list(emb[0, :]))
print(imagels)
uu = str(uuid.uuid1())
name = 'hbw'
sqllite.saveImage(uu, imagels, name)
param = uu
return param
def imageExist(emb):
flag = '-1'
imbs = sqllite.queryImages()
print(imbs)
scoreLs = list()
with ThreadPoolExecutor(8) as executor:
for imb in imbs:
future = executor.submit(cal_compare, emb, imb, scoreLs)
print("result", future.result())
a = future.result()
if a and a.__len__() > 0:
flag = a[0]
break
# t = threading.Thread(target=cal_compare, args=(emb, imb, scoreLs))
# t.start()
# t.join()
return flag
def cal_compare(emb, imb, scoreLs):
em = str2array(imb.image)
score = np.sqrt(np.sum(np.square(np.subtract(emb[0, :], em))))
if score < 0.8:
scoreLs.append(imb.id)
return scoreLs
def getEmb(image):
minsize = 20 # minimum size of face
threshold = [0.6, 0.7, 0.7] # three steps's threshold
factor = 0.709 # scale factor
margin = 44
image_size = 160
print('Creating networks and loading parameters')
img = misc.imread(os.path.expanduser(image), mode='RGB')
img_size = np.asarray(img.shape)[0:2]
bounding_boxes, _ = align.detect_face.detect_face(img, minsize, pnet, rnet, onet, threshold, factor)
img_list = list()
if len(bounding_boxes) >= 1:
det = np.squeeze(bounding_boxes[0, 0:4])
bb = np.zeros(4, dtype=np.int32)
bb[0] = np.maximum(det[0] - margin / 2, 0)
bb[1] = np.maximum(det[1] - margin / 2, 0)
bb[2] = np.minimum(det[2] + margin / 2, img_size[1])
bb[3] = np.minimum(det[3] + margin / 2, img_size[0])
cropped = img[bb[1]:bb[3], bb[0]:bb[2], :]
aligned = misc.imresize(cropped, (image_size, image_size), interp='bilinear')
prewhitened = facenet.prewhiten(aligned)
img_list.append(prewhitened)
images = np.stack(img_list)
feed_dict = {images_placeholder: images, phase_train_placeholder: False}
emb = sess.run(embeddings, feed_dict=feed_dict)
return emb
return '-1'
def str2array(s):
s = s.replace('[', '')
s = s.replace(']', '')
ss = s.split(',')
ls = list()
for i in ss:
ls.append(i)
k = list(map(float, ls))
na = np.array(k)
return na
#拍照检测
# def cutTest(image):
# minsize = 20 # minimum size of face
# threshold = [0.6, 0.7, 0.7] # three steps's threshold
# factor = 0.709 # scale factor
# print('Creating networks and loading parameters')
# img = misc.imread(os.path.expanduser(image), mode='RGB')
# bounding_boxes, points = align.detect_face.detect_face(img, minsize, pnet, rnet, onet, threshold, factor)
#
# eye1_w = points[0]
# eye2_w = points[1]
# nose_w = points[2]
# mouth1_w = points[3]
# mouth2_w = points[4]
# eye1_h = points[5]
# eye2_h = points[6]
# nose_h = points[7]
# mouth1_h = points[8]
# mouth2_h = points[9]
#
# eye_len = cal_len(eye1_w, eye1_h, eye2_w, eye2_h)
# mouth_len = cal_len(mouth1_w, mouth1_h, mouth2_w, mouth2_h)
# left_eye_nose_len = cal_len(eye1_w, eye1_h, nose_w, nose_h)
# right_eye_nose_len = cal_len(eye2_w, eye2_h, nose_w, nose_h)
# left_mouth_nose_len = cal_len(mouth1_w, mouth1_h, nose_w, nose_h)
# right_mouth_nose_len = cal_len(mouth2_w, mouth2_h, nose_w, nose_h)
# eye_nose_len = cal_len((eye1_w+eye2_w)/2, (eye1_h+eye2_h)/2, nose_w, nose_h)
# mouth_nose_len = cal_len((mouth1_w+mouth2_w)/2, (mouth1_h+mouth2_h)/2, nose_w, nose_h)
#
# print("眼睛距离", eye_len)
# print("嘴间距离", mouth_len)
# print("左眼鼻距离", left_eye_nose_len)
# print("右眼鼻距离", right_eye_nose_len)
# print("左嘴鼻距离", left_mouth_nose_len)
# print("右嘴鼻距离", right_mouth_nose_len)
#
# param1 = round(eye_len/mouth_len, 3)
# param2 = round((right_eye_nose_len+left_eye_nose_len)/(right_mouth_nose_len+left_mouth_nose_len), 3)
# param3 = round(eye_nose_len/mouth_nose_len, 3)
#
# a = cal_dif(300, param1, param2, param3)
# param = round(np.log10(a), 3)
# return param
def cal_len(x1, y1, x2, y2):
eye_em = np.array([[x1, y1], [x2, y2]])
eye_len = np.sqrt(np.sum(np.square(np.subtract(eye_em[0, :], eye_em[1, :]))))
return eye_len
def opt(image, minsize, threshold, factor, margin, image_size, img_list, image_paths):
img = misc.imread(os.path.expanduser(image), mode='RGB')
img_size = np.asarray(img.shape)[0:2]
bounding_boxes, _ = align.detect_face.detect_face(img, minsize, pnet, rnet, onet, threshold, factor)
if len(bounding_boxes) >= 1:
det = np.squeeze(bounding_boxes[0, 0:4])
bb = np.zeros(4, dtype=np.int32)
bb[0] = np.maximum(det[0] - margin / 2, 0)
bb[1] = np.maximum(det[1] - margin / 2, 0)
bb[2] = np.minimum(det[2] + margin / 2, img_size[1])
bb[3] = np.minimum(det[3] + margin / 2, img_size[0])
cropped = img[bb[1]:bb[3], bb[0]:bb[2], :]
aligned = misc.imresize(cropped, (image_size, image_size), interp='bilinear')
prewhitened = facenet.prewhiten(aligned)
img_list.append(prewhitened)
else:
image_paths.remove(image)
print("can't detect face, remove ", image)
def load_and_align_data(image_paths, image_size, margin):
minsize = 20 # minimum size of face
threshold = [0.6, 0.7, 0.7] # three steps's threshold
factor = 0.709 # scale factor
print('Creating networks and loading parameters')
tmp_image_paths = copy.copy(image_paths)
img_list = []
with ThreadPoolExecutor(8) as executor:
for image in tmp_image_paths:
executor.submit(opt, image, minsize, threshold, factor, margin, image_size, img_list, image_paths)
# t = threading.Thread(target=opt, args=(image, minsize, threshold, factor, margin, image_size, img_list
# , image_paths))
# t.start()
# t.join()
images = np.stack(img_list)
return images
def parse_arguments(argv):
parser = argparse.ArgumentParser()
parser.add_argument('image_files', type=str, nargs='+', help='Images to compare')
parser.add_argument('--image_size', type=int,
help='Image size (height, width) in pixels.', default=160)
parser.add_argument('--margin', type=int,
help='Margin for the crop around the bounding box (height, width) in pixels.', default=44)
parser.add_argument('--gpu_memory_fraction', type=float,
help='Upper bound on the amount of GPU memory that will be used by the process.', default=1.0)
return parser.parse_args(argv)
def parse_arguments2(argv):
parser = argparse.ArgumentParser()
parser.add_argument('model', type=str,
help='Could be either a directory containing the meta_file and ckpt_file or a model protobuf (.pb) file')
parser.add_argument('--image_size', type=int,
help='Image size (height, width) in pixels.', default=160)
parser.add_argument('--margin', type=int,
help='Margin for the crop around the bounding box (height, width) in pixels.', default=44)
parser.add_argument('--gpu_memory_fraction', type=float,
help='Upper bound on the amount of GPU memory that will be used by the process.', default=1.0)
return parser.parse_args(argv)
app = Flask(__name__)
def resize2(filepath):
img = cv2.imread(filepath)
img2 = cv2.resize(img, (160, 160), interpolation=cv2.INTER_AREA)
cv2.imwrite(filepath, img2)
def writeImage(imPath, im):
with open(imPath, "wb") as f:
f.write(im)
#图片对比接口
@app.route("/compare", methods=["POST"])
def compare():
#获取图片
dic = json.loads(request.get_data())
image1 = dic["image1"]
# image2 = dic["image2"]
im1 = base64.b64decode(image1)
# im2 = base64.b64decode(image2)
im1path = "C:/Users/hbw/facenet/src/img/"+str(uuid.uuid1())+".png"
# im2path = "C:/Users/hbw/facenet/src/img/"+str(uuid.uuid1())+".png"
#写入并处理图片
writeImage(im1path, im1)
# writeImage(im2path, im2)
resize2(im1path)
# resize2(im2path)
#调用对比方法
# imls = list()
# imls.append(im1path)
# imls.append(im2path)
t = dict()
t['data'] = main(im1path)
#删除图片
os.remove(im1path)
# os.remove(im2path)
#返回结果
return json.dumps(t, ensure_ascii=False)
#图片检测接口
@app.route("/detect", methods=["post"])
def detect():
#获取图片
dic = json.loads(request.get_data())
image1 = dic["image1"]
im1 = base64.b64decode(image1)
im1path = "C:/Users/hbw/facenet/src/img/"+str(uuid.uuid1())+".png"
#写入图片
writeImage(im1path, im1)
# resize2(im1path)
#调用检测方法
# flag = dic["flag"]
t = dict()
t["data"] = str(test(im1path, "1"))
#删除图片
os.remove(im1path)
#返回结果
return json.dumps(t, ensure_ascii=False)
#拍照功能
@app.route("/cut", methods=["post"])
def cut():
# 获取图片
dic = json.loads(request.get_data())
image1 = dic["image1"]
im1 = base64.b64decode(image1)
im1path = "C:/Users/hbw/facenet/src/img/" + str(uuid.uuid1()) + ".png"
# 写入图片
writeImage(im1path, im1)
resize2(im1path)
# 调用检测方法
# imls = list()
# imls.append(im1path)
param = cutTest(im1path)
print("param", param)
t = dict()
t["data"] = param
# 删除图片
os.remove(im1path)
# 返回结果
return json.dumps(t, ensure_ascii=False)
@app.route("/deleteall", methods=["post"])
def deleteAll():
sqllite.removeImageAll()
t = dict()
t['data'] = '1'
return json.dumps(t, ensure_ascii=False)
if __name__ == '__main__':
#设置使用GPU
os.environ["CUDA_VISIBLE_DEVICES"] = "0"
config = tf.ConfigProto()
config.gpu_options.per_process_gpu_memory_fraction = 0.7 # 占用GPU90%的显存
config.gpu_options.allow_growth = True
session = tf.Session(config=config)
#设置模型并运行
models = list()
models.append('C:/Users/hbw/facenet/src/20180402-114759/20180402-114759.pb')
args = parse_arguments2(models)
with tf.Graph().as_default():
gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=args.gpu_memory_fraction)
sess = tf.Session(config=tf.ConfigProto(gpu_options=gpu_options, log_device_placement=False))
with sess.as_default():
pnet, rnet, onet = align.detect_face.create_mtcnn(sess, None)
# with tf.Session() as sess:
# Load the model
facenet.load_model(args.model)
# Get input and output tensors
images_placeholder = tf.get_default_graph().get_tensor_by_name("input:0")
embeddings = tf.get_default_graph().get_tensor_by_name("embeddings:0")
phase_train_placeholder = tf.get_default_graph().get_tensor_by_name("phase_train:0")
#人脸检测dblib加载
detector = dlib.get_frontal_face_detector()
predictor = dlib.shape_predictor('E:/dev/facerecognize/blib/shape_predictor_68_face_landmarks.dat')
app.run(host='192.168.2.171', port=8000, debug=False, threaded=True)
sql：
import sqlite3
def saveImage(id, image, name):
sql = 'insert into image (id, image, name) values ("'+id+'", "'+image+'", "'+name+'")'
optTable(sql)
def queryImages():
sql = 'select id, image, name from image'
re = optTable2(sql)
return re
def removeImageAll():
sql = 'delete from image'
optTable(sql)
def optTable(sql):
conn = sqlite3.connect('test.db', check_same_thread=False)
try:
cursor = conn.cursor()
cursor.execute(sql)
cursor.close()
conn.commit()
finally:
conn.close()
def optTable2(sql):
conn = sqlite3.connect('test.db', check_same_thread=False)
ims = list()
try:
cursor = conn.cursor()
re = cursor.execute(sql)
print("re", re)
for row in re:
imInfo = ImageInfo(row[0], row[1], row[2])
ims.append(imInfo)
print(row[0], row[1], row[2])
return ims
finally:
conn.close()
class ImageInfo(object):
def __init__(self, id, image, name):
self.id = id
self.image = image
self.name = name
if __name__ == "__main__":
print("main")
ls = queryImages()
for im in ls:
print(im.id)
print(im.image)
print(im.name)
