機械学習で乃木坂46を顏分類してみた

import urllib.request
from urllib.parse import quote
import httplib2
import json
import os
import cv2
import sys
import shutil
#keywordsの画像のurlを取得後、jpg画像に変換しファイルにどんどん入れてく
#全５人100個ずつ取得
API_KEY = ""#省略
CUSTOM_SEARCH_ENGINE = ""#省略
keywords=["生田絵梨花","齋藤飛鳥","白石麻衣","西野七瀬","橋本奈々未"]
def get_image_url(search_item, total_num):
img_list = []
i = 0
while i < total_num:
query_img = "https://www.googleapis.com/customsearch/v1?key=" + API_KEY + "&cx=" + CUSTOM_SEARCH_ENGINE + "&num=" + str(10 if(total_num-i)>10 else (total_num-i)) + "&start=" + str(i+1) + "&q=" + quote(search_item) + "&searchType=image"
res = urllib.request.urlopen(query_img)
data = json.loads(res.read().decode('utf-8'))
for j in range(len(data["items"])):
img_list.append(data["items"][j]["link"])
i += 10
return img_list
def get_image(search_item, img_list,j):
opener = urllib.request.build_opener()
http = httplib2.Http(".cache")
for i in range(len(img_list)):
try:
fn, ext = os.path.splitext(img_list[i])
print(img_list[i])
response, content = http.request(img_list[i])
filename = os.path.join("./origin_image",str("{0:02d}".format(j))+"."+str(i)+".jpg")
with open(filename, 'wb') as f:
f.write(content)
except:
print("failed to download the image.")
continue
for j in range(len(keywords)):
print(keywords[j])
img_list = get_image_url(keywords[j],100)
get_image(keywords[j], img_list,j)

import numpy as np
import cv2
import matplotlib.pyplot as plt
import numpy as np
from PIL import Image
import glob
import os
#元画像を取り出して顔部分を正方形で囲み、64×64pにリサイズ、別のファイルにどんどん入れてく
in_dir = "./origin_image/*"
out_dir = "./face_image"
in_jpg=glob.glob(in_dir)
in_fileName=os.listdir("./origin_image/")
# print(in_jpg)
# print(in_fileName)
print(len(in_jpg))
for num in range(len(in_jpg)):
image=cv2.imread(str(in_jpg[num]))
if image is None:
print("Not open:",line)
continue
image_gs = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
cascade = cv2.CascadeClassifier("/usr/local/opt/opencv/share/OpenCV/haarcascades/haarcascade_frontalface_alt.xml")
# 顔認識の実行
face_list=cascade.detectMultiScale(image_gs, scaleFactor=1.1, minNeighbors=2,minSize=(64,64))
#顔が１つ以上検出された時
if len(face_list) > 0:
for rect in face_list:
x,y,width,height=rect
image = image[rect[1]:rect[1]+rect[3],rect[0]:rect[0]+rect[2]] if image.shape[0]<64:
continue
image = cv2.resize(image,(64,64))
#顔が検出されなかった時
else:
print("no face")
continue
print(image.shape)
#保存
fileName=os.path.join(out_dir,str(in_fileName[num])+".jpg")
cv2.imwrite(str(fileName),image)
in_dir = "./face_image/*"
in_jpg=glob.glob(in_dir)
img_file_name_list=os.listdir("./face_image/")
#img_file_name_listをシャッフル、そのうち2割をtest_imageディテクトリに入れる
random.shuffle(in_jpg)
import shutil
for i in range(len(in_jpg)//5):
shutil.move(str(in_jpg[i]), "./test_image")

#左右反転、閾値処理、ぼかしで８倍の水増し
import os
import numpy as np
import matplotlib.pyplot as plt
import cv2
def scratch_image(img, flip=True, thr=True, filt=True):
# 水増しの手法を配列にまとめる
methods = [flip, thr, filt]
# ぼかしに使うフィルターの作成
filter1 = np.ones((3, 3))
# オリジナルの画像データを配列に格納
images = [img]
# 手法に用いる関数
scratch = np.array([
lambda x: cv2.flip(x, 1),
lambda x: cv2.threshold(x, 100, 255, cv2.THRESH_TOZERO)[1],
lambda x: cv2.GaussianBlur(x, (5, 5), 0),
])
# 加工した画像を元と合わせて水増し
doubling_images = lambda f, imag: np.r_[imag, [f(i) for i in imag]]
for func in scratch[methods]:
images = doubling_images(func, images)
return images
# 画像の読み込み
in_dir = "./face_image/*"
in_jpg=glob.glob(in_dir)
img_file_name_list=os.listdir("./face_image/")
for i in range(len(in_jpg)):
print(str(in_jpg[i]))
img = cv2.imread(str(in_jpg[i]))
scratch_face_images = scratch_image(img)
for num, im in enumerate(scratch_face_images):
fn, ext = os.path.splitext(img_file_name_list[i])
file_name=os.path.join("./face_scratch_image"[f:id:shintarom4869:20171220032312p:plain],str(fn+"."+str(num)+".jpg"))
cv2.imwrite(str(file_name) ,im)

# 画像と正解ラベルをリストにする
import random from keras.utils.np_utils
​​​​​​​import to_categorical
import os
import cv2
import numpy as np
import matplotlib.pyplot as plt
img_file_name_list=os.listdir("./face_scratch_image/")
print(len(img_file_name_list))
for i in range(len(img_file_name_list)):
n=os.path.join("./face_scratch_image",img_file_name_list[i])
img = cv2.imread(n)
if isinstance(img,type(None)) == True:
img_file_name_list.pop(i)
continue
print(len(img_file_name_list))
X_train=[]
y_train=[]
for j in range(0,len(img_file_name_list)-1):
n=os.path.join("./face_scratch_image/",img_file_name_list[j])
img = cv2.imread(n)
b,g,r = cv2.split(img)
img = cv2.merge([r,g,b])
X_train.append(img)
n=img_file_name_list[j]
y_train=np.append(y_train,int(n[0:2])).reshape(j+1,1)
X_train=np.array(X_train)
img_file_name_list=os.listdir("./test_image1/")
print(len(img_file_name_list))
for i in range(len(img_file_name_list)):
n=os.path.join("./test_image1",img_file_name_list[i])
img = cv2.imread(n)
if isinstance(img,type(None)) == True:
img_file_name_list.pop(i)
continue
print(len(img_file_name_list))
X_test=[]
y_test=[]
for j in range(0,len(img_file_name_list)):
n=os.path.join("./test_image1",img_file_name_list[j])
img = cv2.imread(n)
b,g,r = cv2.split(img)
img = cv2.merge([r,g,b])
X_test.append(img)
n=img_file_name_list[j]
y_test=np.append(y_test,int(n[0:2])).reshape(j+1,1)
X_test=np.array(X_test)​​​​​​​

- 入力 (64x64 3chカラー)
- 畳み込み層1
- プーリング層1
- 畳み込み層2
- プーリング層2
- 畳み込み層3
- プーリング層3
- 平坦層1
- 全結合層1(sigmoid関数)
- 全結合層2(sigmoid関数)
- 全結合層3(softmax関数 出力5)

from keras.layers import Activation, Conv2D, Dense, Flatten, MaxPooling2D
from keras.models import Sequential, load_model
y_train = to_categorical(y_train)
y_test = to_categorical(y_test)
#plt.imshow(X_train[0])
#plt.show()
#print(y_train[0])
# モデルの定義
model = Sequential()
model.add(Conv2D(input_shape=(64, 64, 3), filters=32,kernel_size=(2, 2), strides=(1, 1), padding="same"))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Conv2D(filters=32, kernel_size=(2, 2), strides=(1, 1), padding="same"))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Conv2D(filters=32, kernel_size=(2, 2), strides=(1, 1), padding="same"))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Flatten())
model.add(Dense(256))
model.add(Activation("sigmoid"))
model.add(Dense(128))
model.add(Activation('sigmoid'))
model.add(Dense(5))
model.add(Activation('softmax'))
# コンパイル
model.compile(optimizer='sgd', loss='categorical_crossentropy',metrics=['accuracy'])
# 学習
# model.fit(X_train, y_train, batch_size=32, epochs=50)
#グラフ用
history = model.fit(X_train, y_train, batch_size=32, epochs=100, verbose=1, validation_data=(X_test, y_test))
# 汎化制度の評価・表示
score = model.evaluate(X_test, y_test, batch_size=32, verbose=0)
print('validation loss:{0[0]}\nvalidation accuracy:{0[1]}'.format(score))
#acc, val_accのプロット
plt.plot(history.history["acc"], label="acc", ls="-", marker="o")
plt.plot(history.history["val_acc"], label="val_acc", ls="-", marker="x")
plt.ylabel("accuracy")
plt.xlabel("epoch")
plt.legend(loc="best")
plt.show()
#モデルを保存
model.save("my_model.h5")

import numpy as np
import matplotlib.pyplot as plt
def detect_face(image):
print(image.shape)
#opencvを使って顔抽出
image_gs = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
cascade = cv2.CascadeClassifier("/usr/local/opt/opencv/share/OpenCV/haarcascades/haarcascade_frontalface_alt.xml")
# 顔認識の実行
face_list=cascade.detectMultiScale(image_gs, scaleFactor=1.1, minNeighbors=2,minSize=(64,64))
#顔が１つ以上検出された時
if len(face_list) > 0:
for rect in face_list:
x,y,width,height=rect
cv2.rectangle(image, tuple(rect[0:2]), tuple(rect[0:2]+rect[2:4]), (255, 0, 0), thickness=3)
img = image[rect[1]:rect[1]+rect[3],rect[0]:rect[0]+rect[2]]
if image.shape[0]<64:
print("too small")
continue
img = cv2.resize(image,(64,64))
img=np.expand_dims(img,axis=0)
name = detect_who(img)
cv2.putText(image,name,(x,y+height+20),cv2.FONT_HERSHEY_DUPLEX,1,(255,0,0),2)
#顔が検出されなかった時
else:
print("no face")
return image
def detect_who(img):
#予測
name=""
print(model.predict(img))
nameNumLabel=np.argmax(model.predict(img))
if nameNumLabel== 0:
name="Ikuta Erika"
elif nameNumLabel==1:
name="Saito Asuka"
elif nameNumLabel==2:
name="Shiraishi Mai"
elif nameNumLabel==3:
name="Nishino Nanase"
elif nameNumLabel==4:
name="Hashimoto Nanami"
return name
# model = load_model('./my_model.h5')
image=cv2.imread("./origin_image/01.0.jpg")
if image is None:
print("Not open:")
b,g,r = cv2.split(image)
image = cv2.merge([r,g,b])
whoImage=detect_face(image)
plt.imshow(whoImage)
plt.show()