用 keras 建立超简单的汉字识别模型

    $ wget http://www.nlpr.ia.ac.cn/databases/download/feature_data/HWDB1.1trn_gnt.zip
    # zip 解压没得说, 之后还要解压 alz 压缩文件
    $ wget http://www.nlpr.ia.ac.cn/databases/download/feature_data/HWDB1.1tst_gnt.zip

    import os
    import numpy as np
    import struct
    from PIL import Image


    data_dir = './hanwriting'
    train_data_dir = os.path.join(data_dir, 'HWDB1.1trn_gnt')
    test_data_dir = os.path.join(data_dir, 'HWDB1.1tst_gnt')

    # f = open('3500.txt', 'r', encoding="utf8")
    f = open('3500.txt', 'r')
    total_words = f.readlines()[0].decode("utf-8")
    print(total_words)

    def read_from_gnt_dir(gnt_dir=train_data_dir):
      def one_file(f):
        header_size = 10
        while True:
          header = np.fromfile(f, dtype='uint8', count=header_size)
          if not header.size:   
            break
          sample_size = header[0]   (header[1] << 8)   (header[2] << 16)   (header[3] << 24)
          tagcode = header[5]   (header[4] << 8)
          width = header[6]   (header[7] << 8)
          height = header[8]   (header[9] << 8)
          if header_size   width*height != sample_size:
            break
          image = np.fromfile(f, dtype='uint8', count=width*height).reshape((height, width))
          yield image, tagcode
      for file_name in os.listdir(gnt_dir):
        if file_name.endswith('.gnt'):
          file_path = os.path.join(gnt_dir, file_name)
          with open(file_path, 'rb') as f:
          for image, tagcode in one_file(f):
            yield image, tagcode
    char_set = set()
    for _, tagcode in read_from_gnt_dir(gnt_dir=train_data_dir):
    tagcode_unicode = struct.pack('>H', tagcode).decode('gb2312')
    char_set.add(tagcode_unicode)
    char_list = list(char_set)
    char_dict = dict(zip(sorted(char_list), range(len(char_list))))
    print len(char_dict)  
    import pickle
    f = open('char_dict', 'wb')
    pickle.dump(char_dict, f)
    f.close()
    train_counter = 0
    test_counter = 0

    for image, tagcode in read_from_gnt_dir(gnt_dir=train_data_dir):
      tagcode_unicode = struct.pack('>H', tagcode).decode('gb2312')
      if tagcode_unicode in total_words:
        im = Image.fromarray(image)
        dir_name = './data/train/'   '%0.5d'%char_dict[tagcode_unicode]
        if not os.path.exists(dir_name):
          os.mkdir(dir_name)
        im.convert('RGB').save(dir_name '/'   str(train_counter)   '.png')
        train_counter  = 1
    for image, tagcode in read_from_gnt_dir(gnt_dir=test_data_dir):
      tagcode_unicode = struct.pack('>H', tagcode).decode('gb2312')
      if tagcode_unicode in total_words:
        im = Image.fromarray(image)
        dir_name = './data/test/'   '%0.5d'%char_dict[tagcode_unicode]
        if not os.path.exists(dir_name):
          os.mkdir(dir_name)
        im.convert('RGB').save(dir_name '/'   str(test_counter)   '.png')
        test_counter  = 1

    from __future__ import print_function
    import os
    from keras.preprocessing.image import ImageDataGenerator
    from keras.layers import Input, Dense, Dropout, Convolution2D, MaxPooling2D, Flatten
    from keras.models import Model, load_model

    data_dir = './data'
    train_data_dir = os.path.join(data_dir, 'train')
    test_data_dir = os.path.join(data_dir, 'test')

    # dimensions of our images.
    img_width, img_height = 64, 64
    charset_size = 3751
    nb_validation_samples = 800
    nb_samples_per_epoch = 2000
    nb_nb_epoch = 20000;

    def train(model):
      train_datagen = ImageDataGenerator(
        rescale=1. / 255,
        rotation_range=0,
        width_shift_range=0.1,
        height_shift_range=0.1
      )
      test_datagen = ImageDataGenerator(rescale=1./255)

      train_generator = train_datagen.flow_from_directory(
        train_data_dir,
        target_size=(img_width, img_height),
        batch_size=1024,
        color_mode="grayscale",
        class_mode='categorical')
      validation_generator = test_datagen.flow_from_directory(
        test_data_dir,
        target_size=(img_width, img_height),
        batch_size=1024,
        color_mode="grayscale",
        class_mode='categorical')

      model.compile(loss='categorical_crossentropy',
        optimizer='rmsprop',
        metrics=['accuracy'])
      model.fit_generator(train_generator,
        samples_per_epoch=nb_samples_per_epoch,
        nb_epoch=nb_nb_epoch,
        validation_data=validation_generator,
        nb_val_samples=nb_validation_samples)

    def build_model(include_top=True, input_shape=(64, 64, 1), classes=charset_size):
      img_input = Input(shape=input_shape)
      x = Convolution2D(32, 3, 3, activation='relu', border_mode='same', name='block1_conv1')(img_input)
      x = Convolution2D(32, 3, 3, activation='relu', border_mode='same', name='block1_conv2')(x)
      x = MaxPooling2D((2, 2), strides=(2, 2), name='block1_pool')(x)
      x = Convolution2D(64, 3, 3, activation='relu', border_mode='same', name='block2_conv1')(x)
      x = Convolution2D(64, 3, 3, activation='relu', border_mode='same', name='block2_conv2')(x)
      x = MaxPooling2D((2, 2), strides=(2, 2), name='block2_pool')(x)

      if include_top:
        x = Flatten(name='flatten')(x)
        x = Dropout(0.05)(x)
        x = Dense(1024, activation='relu', name='fc2')(x)
        x = Dense(classes, activation='softmax', name='predictions')(x)

      model = Model(img_input, x, name='model')
      return model

    model = build_model()
    # model = load_model("./model.h5")
    train(model)
    # model.save("./model.h5")