+
+
+
+
+
+
+
+
+
+
+下载数据集¶
数据集来自 http://www.ee.surrey.ac.uk/CVSSP/demos/chars74k/
+我们下载的是EnglishFnt.tgz，是印刷体数字加大小写字母。
+ + +
+
+
+
+
+
+In [1]:
+
+
+
+
+
+
+import requests
+from tqdm import tqdm
+import os
+
+fileurl = 'http://www.ee.surrey.ac.uk/CVSSP/demos/chars74k/EnglishFnt.tgz'
+filename = 'EnglishFnt.tgz'
+if not os.path.exists(filename):
+ r = requests.get(fileurl, stream=True)
+ with open(filename, 'wb') as f:
+ for chunk in tqdm(r.iter_content(1024), unit='KB', total=int(r.headers['Content-Length'])/1024):
+ f.write(chunk)
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+In [2]:
+
+
+
+
+
+
+import tarfile
+import shutil
+
+def mkdir(path):
+ if not os.path.exists(path):
+ os.makedirs(path)
+
+def rmdir(path):
+ if os.path.exists(path):
+ shutil.rmtree(path)
+
+with tarfile.open(filename, 'r') as tfile:
+ print 'loading'
+ members = tfile.getmembers()
+ for member in tqdm(members):
+ if tarfile.TarInfo.isdir(member):
+ mkdir(member.name)
+ continue
+ with open(member.name, 'wb') as f:
+ f.write(tfile.extractfile(member).read())
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+In [3]:
+
+
+
+
+
+
+notnumdir = 'English/Fnt/Sample011/'
+for i in tqdm(range(12, 63)):
+ path = 'English/Fnt/Sample%03d/' % i
+ for filename in os.listdir(path):
+ os.rename(path+filename, notnumdir+filename)
+ os.rmdir(path)
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+In [4]:
+
+
+
+
+
+
+import cv2
+import numpy as np
+
+def resize(rawimg): # resize img to 28*28
+ fx = 28.0 / rawimg.shape[0]
+ fy = 28.0 / rawimg.shape[1]
+ fx = fy = min(fx, fy)
+ img = cv2.resize(rawimg, None, fx=fx, fy=fy, interpolation=cv2.INTER_CUBIC)
+ outimg = np.ones((28, 28), dtype=np.uint8) * 255
+ w = img.shape[1]
+ h = img.shape[0]
+ x = (28 - w) / 2
+ y = (28 - h) / 2
+ outimg[y:y+h, x:x+w] = img
+ return outimg
+
+def convert(imgpath):
+ img = cv2.imread(imgpath)
+ gray = cv2.imread(imgpath, cv2.IMREAD_GRAYSCALE)
+ bw = cv2.adaptiveThreshold(gray, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY_INV, 25, 25)
+ img2, ctrs, hier = cv2.findContours(bw.copy(), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
+ rects = [cv2.boundingRect(ctr) for ctr in ctrs]
+ x, y, w, h = rects[-1]
+ roi = gray[y:y+h, x:x+w]
+ return resize(roi)
+
+
+
+
+
+In [5]:
+
+
+
+
+
+
+import matplotlib.pyplot as plt
+
+%matplotlib inline
+
+imgpath = 'English/Fnt/Sample001/img001-00001.png'
+img = cv2.imread(imgpath)
+rsz = convert(imgpath)
+
+plt.subplot(1, 2, 1)
+plt.imshow(img, cmap='gray')
+plt.subplot(1, 2, 2)
+plt.imshow(rsz, cmap='gray')
+
+
+
+
+
+
+
+
+
+Out[5]:
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+预处理所有图片¶
+
+
+
+
+
+In [6]:
+
+
+
+
+
+
+rmdir('train')
+
+for i in range(11):
+ path = 'English/Fnt/Sample%03d/' % (i+1)
+ trainpath = 'train/%d/' % i
+ mkdir(trainpath)
+ for filename in tqdm(os.listdir(path), desc=trainpath):
+ try:
+ cv2.imwrite(trainpath + filename, convert(path + filename))
+ except:
+ pass
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+In [11]:
+
+
+
+
+
+
+from sklearn.model_selection import train_test_split
+for i in range(11):
+ trainpath = 'train/%d/' % i
+ validpath = 'valid/%d/' % i
+ mkdir(validpath)
+ imgs = os.listdir(trainpath)
+ trainimgs, validimgs = train_test_split(imgs, test_size=0.1)
+ for filename in validimgs:
+ os.rename(trainpath+filename, validpath+filename)
+
+
+
+
+
+In [ ]:
+
+
+
+
+
+
+
+
+
+