keras-文本图片文字识别

1. Keras环境安装

##参考Keras安装点击打开链接 <http://keras-cn.readthedocs.io/en/latest/#_2>

2. 文本图片素材－文字切割并保存切割图片
# -*- coding: UTF-8 -*- import cv2 import numpy as np import matplotlib.pyplot
as plt def median_split_ranges(peek_ranges): new_peek_ranges = [] widthes = []
for peek_range in peek_ranges: w = peek_range[1] - peek_range[0] + 1
widthes.append(w) widthes = np.asarray(widthes) median_w = np.median(widthes)
for i, peek_range in enumerate(peek_ranges): num_char =
int(round(widthes[i]/median_w, 0)) if num_char > 1: char_w = float(widthes[i] /
num_char) for i in range(num_char): start_point = peek_range[0] + int(i *
char_w) end_point = peek_range[0] + int((i + 1) * char_w)
new_peek_ranges.append((start_point, end_point)) else:
new_peek_ranges.append(peek_range) return new_peek_ranges def
extract_peek_ranges_from_array(array_vals, minimun_val=10, minimun_range=2):
start_i = None end_i = None peek_ranges = [] for i, val in
enumerate(array_vals): if val > minimun_val and start_i is None: start_i = i
elif val > minimun_val and start_i is not None: pass elif val < minimun_val and
start_i is not None: end_i = i if end_i - start_i >= minimun_range:
peek_ranges.append((start_i, end_i)) start_i = None end_i = None elif val <
minimun_val and start_i is None: pass else: raise ValueError("cannot parse this
case...") return peek_ranges def get_font_face_peek_ranges(path_test_image):
image_color = cv2.imread(path_test_image) new_shape = (image_color.shape[1] *
2, image_color.shape[0] * 2) image_color = cv2.resize(image_color, new_shape)
image = cv2.cvtColor(image_color, cv2.COLOR_BGR2GRAY) adaptive_threshold =
cv2.adaptiveThreshold( image, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
cv2.THRESH_BINARY_INV, 11, 2) horizontal_sum = np.sum(adaptive_threshold,
axis=1) plt.plot(horizontal_sum, range(horizontal_sum.shape[0]))
plt.gca().invert_yaxis() # plt.show() peek_ranges =
extract_peek_ranges_from_array(horizontal_sum) vertical_peek_ranges2d = [] for
peek_range in peek_ranges: start_y = peek_range[0] end_y = peek_range[1]
line_img = adaptive_threshold[start_y:end_y, :] vertical_sum = np.sum(line_img,
axis=0) vertical_peek_ranges = extract_peek_ranges_from_array( vertical_sum,
minimun_val=40, minimun_range=1)
vertical_peek_ranges2d.append(vertical_peek_ranges) vertical_peek_ranges2d = []
for peek_range in peek_ranges: start_y = peek_range[0] end_y = peek_range[1]
line_img = adaptive_threshold[start_y:end_y, :] vertical_sum = np.sum(line_img,
axis=0) vertical_peek_ranges = extract_peek_ranges_from_array( vertical_sum,
minimun_val=40, minimun_range=1) vertical_peek_ranges =
median_split_ranges(vertical_peek_ranges)
vertical_peek_ranges2d.append(vertical_peek_ranges) return
peek_ranges,vertical_peek_ranges2d,image_color color = (0, 0, 255)
path_test_image = "tmp/font.png" peek_ranges,vertical_peek_ranges2d,image_color
= get_font_face_peek_ranges(path_test_image) for i, peek_range in
enumerate(peek_ranges): for (j,vertical_range) in
enumerate(vertical_peek_ranges2d[i]): x = vertical_range[0] y = peek_range[0] w
= vertical_range[1] - x h = peek_range[1] - y image = image_color[y - 2:y + h +
2, x - 2:x + w + 2] pt1 = (x, y) pt2 = (x + w, y + h)
cv2.rectangle(image_color, pt1, pt2, color) cv2.imshow('image', image_color)
cv2.waitKey(0)
3. 训练
FILE_PATH = "model.h5" #模型进行存储和读取的地方 IMAGE_SIZE = 128 PATH = "fonts"
imgs,labels,counter = read_file(PATH, IMAGE_SIZE) X_train,X_test,y_train,y_test
= train_test_split(imgs,labels,test_size=0.2,random_state=0) X_train =
X_train.reshape(X_train.shape[0], 1, IMAGE_SIZE, IMAGE_SIZE)/255.0 X_test =
X_test.reshape(X_test.shape[0], 1, IMAGE_SIZE, IMAGE_SIZE) / 255.0 X_train =
X_train.astype('float32') X_test = X_test.astype('float32') Y_train =
np_utils.to_categorical(y_train, num_classes=counter) Y_test =
np_utils.to_categorical(y_test, num_classes=counter) model = Sequential()
model.add( Convolution2D( filters=32, kernel_size=(5, 5), padding='same',
dim_ordering='th', input_shape=X_train.shape[1:] ) )
model.add(Activation('relu')) model.add( MaxPooling2D( pool_size=(2, 2),
strides=(2, 2), padding='same' ) ) model.add(Convolution2D(filters=64,
kernel_size=(5, 5), padding='same')) model.add(Activation('relu'))
model.add(MaxPooling2D(pool_size=(2, 2), strides=(2, 2), padding='same'))
model.add(Flatten()) model.add(Dense(512)) model.add(Activation('relu'))
model.add(Dense(counter)) model.add(Activation('softmax')) model.summary()
model.compile( optimizer='adam', optimizer， loss='categorical_crossentropy',
metrics=['accuracy']) model.fit(X_train,Y_train,epochs=32,batch_size=32) loss,
accuracy = model.evaluate(X_test, Y_test) print('test loss;', loss) print('test
accuracy:', accuracy) model.save(FILE_PATH)
4. 识别文字图片

a. 图片文字切割

b. 文字识别
# -*- coding: UTF-8 -*- from keras.models import load_model import cv2 import
numpy as np import utils import os def getLetter(model, img, name_list,
IMAGE_SIZE): img = cv2.resize(img, (IMAGE_SIZE, IMAGE_SIZE)) img =
cv2.cvtColor(img, cv2.COLOR_BGR2GRAY) img = img.reshape((1, 1, IMAGE_SIZE,
IMAGE_SIZE)) img = img.astype('float32') img = img/255.0 result =
model.predict_proba(img) max_index = np.argmax(result) picType,prob =
max_index,result[0][max_index] if picType != -1: return name_list[picType],prob
else: return "" FILE_PATH = "model.h5" IMAGE_SIZE = 128 result = ""
path_test_image = "tmp/font.png" name_list = readName() model =
load_model(FILE_PATH) peek_ranges,vertical_peek_ranges2d,image_color =
utils.get_font_face_peek_ranges(path_test_image) for i, peek_range in
enumerate(peek_ranges): for (j,vertical_range) in
enumerate(vertical_peek_ranges2d[i]): x = vertical_range[0] y = peek_range[0] w
= vertical_range[1] - x h = peek_range[1] - y image = image_color[y - 2:y + h +
2, x - 2:x + w + 2] letter,prob = getLetter(model, image, name_list,
IMAGE_SIZE) code = letter result += code print(result)


c. 测试结果(还需优化）

• « 上一篇：【ML数学知识】极大似然估计
• » 下一篇：pycharm中tensorflow代码不能自动补全或import红线问题解决