本系列文章由 @yhl_leo 出品,转载请注明出处。
文章链接: http://blog.csdn.net/yhl_leo/article/details/52886351
Python code : yhlleo/textRegionMask
根据图像中文本字符的坐标信息,生成文本区域mask图像。如下图
文本字符信息记录格式为:
bjtextset01_0004.jpg
1
1 527.50 243.50 581.67 311.00 "2"
其中,bjtextset01_0004.jpg
为图像名(全小写字符),紧接着的1
为包含文本字符的数量,后面接着就是对应的文本字符的位置坐标527.50 243.50 581.67 311.00
(格式为x, y, x, y
,即两个顶点坐标),2
为字符内容,该行最前面的1
为标记符,可以忽略。
首先,读取文本内容:
import os
import copy as cp
class DataGt(object):
"""docstring for DataGt"""
def __init__(self, fname, trlist):
super(DataGt, self).__init__()
self.fname = fname
self.trlist = trlist
def loaddata(path):
fp = open(path).read().splitlines()
gt = DataGt([],[])
niter = 0
idx = 0
while niter < len(fp):
if '.jpg' in fp[idx]:
textlst = []
gt.fname.append(fp[idx]);
idx = idx + 1
num = int(fp[idx])
for i in range(num):
idx = idx + 1
if '1' in fp[idx] and '"' in fp[idx]:
loc = fp[idx].split(' ')[1:5]
textlst.append(loc)
gt.trlist.append(textlst)
else:
idx = idx + 1
niter = idx
return gt
然后,绘制mask图:
import os
import cv2
import loadgt
import numpy as np
def im_lists( path ):
return os.listdir(path);
def path_insensitive(lst, fn):
for ln in lst:
if ln.lower() == fn.lower():
return ln
return None
def genMask(gt, im_path, savepath):
num = len(gt.fname)
ims = im_lists(im_path)
for idx in range(num):
fn = path_insensitive( ims, gt.fname[idx] )
fname = os.path.join(im_path, fn)
sname = os.path.join(savepath, fn)
im = cv2.imread(fname)
size_im = im.shape
#print size_im
mask = np.zeros([size_im[0], size_im[1]], dtype=np.uint8)
for ls in gt.trlist[idx]:
mask[int(float(ls[1])):int(float(ls[3])), int(float(ls[0])): int(float(ls[2]))] = 255
cv2.imwrite(sname, mask, [cv2.cv.CV_IMWRITE_PNG_COMPRESSION, 0])
im_path = "./data"
savepath = "./mask"
gtpath = "./test.txt"
gt = loadgt.loaddata(gtpath)
genMask(gt,im_path, savepath)
结果如图: