def get_next_batch(batch_size=64, cnt=0):
"""
# 生成一个训练batch
:param batch_size cnt
:return:
"""
batch_x = np.zeros([batch_size, IMAGE_HEIGHT * IMAGE_WIDTH])
batch_y = np.zeros([batch_size, MAX_CAPTCHA * CHAR_SET_LEN])
f = open(root + "mappings.txt", 'r')
lines = f.readlines()
f.close()
i = 0
for j in range(cnt * batch_size, (cnt + 1) * batch_size):
text = lines[j].split(",")[-1]
text = text.split("=")[0]
image = Image.open(root + str(j).zfill(4) + ".jpg")
# print(j)
# print(text)
# image.show()
image = convert2gray(image)
batch_x[
i, :] = image.flatten() / 255 # (image.flatten()-128)/128 mean为0
batch_y[i, :] = text2vec(text)
i += 1
return batch_x, batch_y
def batch_hack_captcha(inroad, outroad):
try:
fw = open(outroad, 'w')
with tf.Session() as sess:
output = crack_captcha_cnn()
saver = tf.train.Saver()
saver.restore(sess, tf.train.latest_checkpoint(model_path))
dirs = os.listdir(inroad)
for i in dirs:
QApplication.processEvents()
image = Image.open(inroad + '/' + i)
# 用于测试测试集准确率
# text = lines[i].split(",")[1].strip()
# text1 = text.split("=")[0]
# label_ans = eval(text1)
image = convert2gray(image)
image = image.flatten() / 255
pred = hack_function(sess, tf.argmax(tf.reshape(output, [-1, MAX_CAPTCHA, CHAR_SET_LEN]), 2), image)
predict_text = eval(pred)
# if label_ans == predict_text:
# right_cnt += 1
# else:
# pass
i = i.split(".")[0]
print("{},{}".format(i, str(pred) + "=" + str(predict_text)))
fw.write("{},{}\n".format(i, str(pred) + "=" + str(predict_text)))
fw.flush()
except:
print("ERROR!")
return -1
def batch_hack_captcha():
"""
批量生成验证码，然后再批量进行识别
:return:
"""
# 定义预测计算图
output = crack_captcha_cnn()
predict = tf.argmax(tf.reshape(output, [-1, MAX_CAPTCHA, CHAR_SET_LEN]), 2)
saver = tf.train.Saver()
with tf.Session() as sess:
#saver = tf.train.import_meta_graph(s_path)
saver.restore(sess, tf.train.latest_checkpoint(model_path))
stime = time.time()
#imgpath="D:/gitrepos/captcha-tensorflow/work/crack/y-capt-data/capt-python-36/train"
imgpath="E:/MyProjects/captcha-tensorflow/vcode1"
imgs=os.listdir(imgpath)
task_cnt = len(imgs)
right_cnt = 0
for i in imgs:
#text, image = wrap_gen_captcha_text_and_image()
text=i.replace('.png','')
img=(imgpath+"/"+i)
#print(text)
#img = tf.read_file(img)
#img = tf.image.decode_png(img)
#img = tf.image.convert_image_dtype(img, dtype=tf.uint8)
img = Image.open(img)
img=img.convert("RGB")
#img=img.resize((160,60),Image.ANTIALIAS)
img=np.array(img)
#print(img)
image = convert2gray(img)
image = image.flatten()/255
predict_text = hack_function(sess, predict, image)
if text == predict_text:
right_cnt += 1
else:
print("标记: {} 预测: {}".format(text, predict_text))
pass
# print("标记: {} 预测: {}".format(text, predict_text))
print('task:', task_cnt, ' cost time:', (time.time() - stime), 's')
print('right/total-----', right_cnt, '/', task_cnt)
print('正确率：',right_cnt/task_cnt)
def recognize(self, bytes):
"""
对验证码图片进行识别。
:param bytes: 图片的二进制数据
:return: 识别结果字符串
"""
try:
bin_img = Image.open(BytesIO(bytes))
except OSError:
self.ui.statusbar.showMessage('输入的数据不是验证码！')
return
if (160, 60) != bin_img.size:
bin_img = bin_img.resize((160, 60))
img = np.array(bin_img)
img = convert2gray(img)
img = img.flatten() / 255
return hack_function(self.sess, self.predict, img)
def get_next_batch(batch_size=128):
"""
# 生成一个训练batch
:param batch_size:
:return:
"""
batch_x = np.zeros([batch_size, IMAGE_HEIGHT * IMAGE_WIDTH])
batch_y = np.zeros([batch_size, MAX_CAPTCHA * CHAR_SET_LEN])
for i in range(batch_size):
text, image = wrap_gen_captcha_text_and_image()
image = convert2gray(image)
batch_x[i, :] = image.flatten() / 255 # (image.flatten()-128)/128 mean为0
batch_y[i, :] = text2vec(text)
return batch_x, batch_y
def predict_captcha():
"""
识别测试集路径的验证码
:return:
"""
output = cnn_architecture.crack_captcha_cnn()
saver = tf.train.Saver()
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
# 获取训练后参数路径
checkpoint = tf.train.get_checkpoint_state("model")
if checkpoint and checkpoint.model_checkpoint_path:
saver.restore(sess, checkpoint.model_checkpoint_path)
print("Successfully loaded:", checkpoint.model_checkpoint_path)
else:
print("Could not find CNN network Model.")
n = 0
test_image_files = os.listdir(config.test_data_path)
for f in test_image_files:
image = Image.open(os.path.join(config.test_data_path, f))
image = np.array(image)
image = utils.convert2gray(image)
image = image.flatten()
time1 = time.time()
predict = tf.argmax(
tf.reshape(output,
[-1, config.MAX_CAPTCHA, config.CHAR_SET_LEN]), 2)
text_list = sess.run(predict,
feed_dict={
cnn_architecture.X: [image],
cnn_architecture.keep_prob: 1
})
predict_text = utils.vec2text(text_list)
time2 = time.time()
elapsed = time2 - time1
print("{} predict:{} elapsed time: {} ms".format(
f, predict_text, format(elapsed * 1000, '0.2f')))
index = f.rfind(".")
if predict_text == f[index - config.MAX_CAPTCHA:index]:
n += 1
print("ACC {}".format(
n / (len(test_image_files) if len(test_image_files) > 0 else 1)))
def get_next_batch(batch_size=128):
"""
# 生成一个训练batch
:param batch_size:
:return:
"""
imgpath = "D:/gitrepos/captcha-tensorflow/vcode1/"
batch_x = np.zeros([batch_size, IMAGE_HEIGHT * IMAGE_WIDTH])
batch_y = np.zeros([batch_size, MAX_CAPTCHA * CHAR_SET_LEN])
td = train_data()
for i in range(batch_size):
#text, image = wrap_gen_captcha_text_and_image()
text, image = td.get_text_img(imgpath)
image = convert2gray(image)
batch_x[
i, :] = image.flatten() / 255 # (image.flatten()-128)/128 mean为0
batch_y[i, :] = text2vec(text)
return batch_x, batch_y
def discern(imgFile):
output = crack_captcha_cnn()
saver = tf.train.Saver()
with tf.Session() as sess:
saver.restore(sess, tf.train.latest_checkpoint(model_path))
for i in range(100):
predict = tf.argmax(
tf.reshape(output, [-1, MAX_CAPTCHA, CHAR_SET_LEN]), 2)
image = Image.open("image/%s.png" % (i + 1))
image = image.resize((160, 60))
image.save("/tmp/big/%s.png" % (i + 1))
image = np.array(image)
image = convert2gray(image)
image = image.flatten() / 255
captcha = hack_function(sess, predict, image)
print("%s:%s" % (i + 1, captcha))
return captcha
def get_captcha(image_path):
"""
识别测试集路径的验证码
:return:
"""
tf.reset_default_graph()
image = Image.open(os.path.join(config.test_data_path, image_path))
image = np.array(image)
image = utils.convert2gray(image)
image = image.flatten()
image = np.reshape(image, (-1, config.IMAGE_HEIGHT, config.IMAGE_WIDTH, 1))
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
# 获取训练后参数路径
checkpoint = tf.train.get_checkpoint_state("model")
if checkpoint and checkpoint.model_checkpoint_path:
saver = tf.train.import_meta_graph(
checkpoint.model_checkpoint_path + '.meta')
saver.restore(sess, checkpoint.model_checkpoint_path)
print("Successfully loaded:", checkpoint.model_checkpoint_path)
graph = tf.get_default_graph() # 加载默认图
input_x = graph.get_tensor_by_name("input_x:0")
out_y = graph.get_tensor_by_name("out_y:0")
keep_prob = graph.get_tensor_by_name("keep_prob:0")
predict = tf.argmax(
tf.reshape(out_y,
[-1, config.MAX_CAPTCHA, config.CHAR_SET_LEN]), 2)
text_list = sess.run(predict,
feed_dict={
input_x: image,
keep_prob: 1
})
predict_text = utils.vec2text(text_list)
return predict_text
else:
print("Could not find CNN network Model.")
return None
def run(self):
right_cnt = 0
for i in range(1, self.count + 1):
if not self.status:
return
text, image = wrap_gen_captcha_text_and_image()
image = convert2gray(image)
image = image.flatten() / 255
predict_text = hack_function(self.sess, self.predict, image)
if text == predict_text:
right_cnt += 1
else:
# print("===========({}/{})\n标记: {}\n预测: {}".format(i, task_cnt, text, predict_text))
self.signal.emit(
QtGui.QListWidget, 'listWidget_2',
"===========({}/{})\n标记: {}\n预测: {}".format(
i, self.count, text, predict_text))
# print("标记: {} 预测: {}".format(text, predict_text))
# print('task:', task_cnt, ' cost time:', (time.time() - stime), 's')
# print('right/total-----', right_cnt, '/', task_cnt)
self.signal.emit(
QtGui.QListWidget, 'listWidget_2',
'right/total-----{}/{}({}%)'.format(right_cnt, self.count,
right_cnt * 100 / self.count))
def get_next_batch(batch_size=128):
"""
# 生成一个训练batch
:param batch_size:
:return:
"""
batch_x = np.zeros([batch_size, IMAGE_HEIGHT * IMAGE_WIDTH])
#生成表示图片的二维向量，也就是输入图片的矩阵？？？？？？
batch_y = np.zeros([batch_size, MAX_CAPTCHA * CHAR_SET_LEN])
#生成一个二维向量，这里是过滤器？？？？？？
for i in range(batch_size):
text, image = wrap_gen_captcha_text_and_image()
image = convert2gray(image)
#x[:,i]表示取所有维中第i个数据，通常返回数组
#x[:,m:n]，即取所有维中第m到n-1个数据，含左不含右
#x[i,:]表示取第一维中下标为i的所有元素，通常返回数组
#flatten() 是将多维数组降位到一维并返回拷贝，默认降维是横向的
batch_x[
i, :] = image.flatten() / 255 # (image.flatten()-128)/128 mean为0
batch_y[i, :] = text2vec(text)
return batch_x, batch_y