SVT（Street View Text）数据集的处理

Street View Text数据集图像来源自于Google Street View，数据集种的图像包含好质量和低质量的图像，通常低质量图片居多。下载的数据集种包含两个文件，train.xml和test.xml文件。格式如下：

<?xml version="1.0" encoding="utf-8"?>
<tagset>
   <image>
      <imageName>img/14_03.jpg</imageName>
      <address>341 Southwest 10th Avenue Portland OR</address>
 <lex>LIVING,ROOM,THEATERS,KENNY,ZUKE,DELICATESSEN,CLYDE,COMMON,ACE,HOTEL,PORTLAND,ROSE,CITY,BOOKS,STUMPTOWN,COFFEE,ROASTERS,RED,CAP,GARAGE,FISH,GROTTO,SEAFOOD,RESTAURANT,AURA,RESTAURANT,LOUNGE,ROCCO,PIZZA,PASTA,BUFFALO,EXCHANGE,MARK,SPENCER,LIGHT,FEZ,BALLROOM,READING,FRENZY,ROXY,SCANDALS,MARTINOTTI,CAFE,DELI,CROWSENBERG,HALF</lex>
      <Resolution x="1280" y="880"/>
      <taggedRectangles>
         <taggedRectangle height="75" width="236" x="375" y="253">
            <tag>LIVING</tag>
         </taggedRectangle>
         <taggedRectangle height="76" width="175" x="639" y="272">
            <tag>ROOM</tag>
         </taggedRectangle>
         <taggedRectangle height="87" width="281" x="839" y="283">
            <tag>THEATERS</tag>
         </taggedRectangle>
      </taggedRectangles>
   </image>

 这种显然不能直接运用，需要对数据集进行处理，程序里包含两个函数read_SVT_dataset和crop函数，前者是要将文字所在的区域信息提取到txt文件中，后者则是从原图像中剪裁出并保存图像。txt文件保存形式如下：

dataset/img/14_03.jpg   # 原图像的路径
3                       # 原图像中识别单词的数目
LIVING,375,253,236,75,  # 识别单词，bounding box[x, y, width, height]
ROOM,639,272,175,76,
THEATERS,839,283,281,87，

程序全部代码如下： 

import os
import sys
import tensorflow as tf
from PIL import Image
from xml.etree import ElementTree

train_XML_src = 'dataset/train.xml'
test_XML_src = 'dataset/test.xml'
image_src = 'dataset/img/'

def read_SVT_dataset(XML_src, write_txt, image_src):
    '''从xml文件中提出图片信息写入txt文件中
    @param XML_src: xml文件路径
    @param write_txt: 提取图片的信息
    @param image_src: 图片文件所在文件夹
    '''
    train_fd = open(write_txt, 'w')
    with open(XML_src) as f:
        tree = ElementTree.parse(f)

    for node in tree.iter('image'):
        img_name = []
        for each_image in node:
            if each_image.tag == 'imageName':
                # 记录保存图像名称
                img_name.append(each_image.text[4::])
                train_fd.write(image_src + each_image.text[4::] +  '\n')

            if each_image.tag == 'taggedRectangles':
                # 计算矩形的数目
                count = 0
                x, y, height, width, name = [], [], [], [], []
                for each_taggedRec in each_image:
                    count = count + 1
                    # 获取坐标信息，得到的是字典
                    tmp_dict = each_taggedRec.attrib
                    x.append(tmp_dict['x'])
                    y.append(tmp_dict['y'])
                    height.append(tmp_dict['height'])
                    width.append(tmp_dict['width'])
                    name.append(each_taggedRec.find('tag').text)

                train_fd.write(str(count) + '\n')
                for i in range(len(x)):
                    train_fd.write(str(name[i]) + ',' + str(x[i]) + ',' + str(y[i]) + ',' + str(width[i]) + ',' +
                                   str(height[i]) + ',' + '\n')

    train_fd.close()


def crop(text):
    '''从txt文本文件中提取信息，剪裁图片
    @param text: txt文件路径
    @return:
    '''
    file = open(text, 'r')
    while True:
        line = file.readline()
        if not line:
            break
        line = line.split('\n')
        image = Image.open(line[0])
        count = int(file.readline())
        for i in range(count):
            string = file.readline()
            list = string.split(',')
            img_name = list[0]
            x, y, width, height = int(list[1]), int(list[2]), int(list[3]), int(list[4])
            crop_img = image.crop([x, y, x + width, y + height])
            crop_img.save('./dataset/images/' + img_name + '.jpg')


def main(argv):
    read_SVT_dataset(train_XML_src, 'train.txt', image_src)
    read_SVT_dataset(test_XML_src, 'test.txt', image_src)
    crop('train.txt')
    crop('test.txt')


if __name__ == '__main__':
    tf.app.run()

最后得到剪裁的图片如下图所示：

总共有350张原图像，得到571张剪裁的图片