cornsea

基于颜色模式的颜色条形码检测(使用python 2.7和opencv 2.2)

这种颜色条形码有3种颜色:青色,粉色和淡黄色。各种颜色区域为矩形,均匀分布在一个黑底的矩形中。

检测的基本思想是:

1) 如果某个像素属于3种颜色的一种,就把对应坐标处标记为1。

  处理完整幅图像后,得到一个2值图像。

2)将1)中得到的2值图像分别向x和y轴投影,也即计算各行和各列的和。

 比如:一行为 101111,向y轴投影后,得到的值是5.

 投影后,得到x,y轴两个向量。

3)在x和y轴向量中搜索最小和最大的非零值,对应的坐标就圈定了条形码所在位置。

  当然这里要设定一个阈值,比如非零值的最小必须大于3或者5之类的,因为矩形颜色块是有最小面积的。

不过,为了能在实际中使用,还需要添加很多处理,比如噪声,颜色变化,视角变化之类的因素也需要考虑。

下面是处理结果的截图,后面有代码:

 

代码
#!/usr/bin/python
#
#

import cv
import sys
import math
names
= ["temp.jpg"];

def findFirstColorPattern(img, pattern):
"""
try to test if one pixel is in our pattern
"""
channels
= [None, None, None]
channels[0]
= cv.CreateImage(cv.GetSize(img),8,1)#blue
channels[1] = cv.CreateImage(cv.GetSize(img),8,1)#green
channels[2] = cv.CreateImage(cv.GetSize(img),8,1)#red
ch0 = cv.CreateImage(cv.GetSize(img),8,1)#blue
ch1 = cv.CreateImage(cv.GetSize(img),8,1)#green
ch2 = cv.CreateImage(cv.GetSize(img),8,1)#red
cv.Split(img, ch0, ch1, ch2, None)
dest0
= cv.CreateImage(cv.GetSize(img),8,1)
dest1
= cv.CreateImage(cv.GetSize(img),8,1)
dest2
= cv.CreateImage(cv.GetSize(img),8,1)
dest3
= cv.CreateImage(cv.GetSize(img),8,1)
cv.Smooth(ch0, channels[0], cv.CV_GAUSSIAN,
3, 3, 0)
cv.Smooth(ch1, channels[
1], cv.CV_GAUSSIAN, 3, 3, 0)
cv.Smooth(ch2, channels[
2], cv.CV_GAUSSIAN, 3, 3, 0)
result
=[]
for i in range(3):
lower
= pattern[i][2] - 25
upper
= pattern[i][2] + 25
cv.InRangeS(channels[0],lower,upper,dest0)
lower
= pattern[i][1] - 25
upper
= pattern[i][1] + 25
cv.InRangeS(channels[
1],lower,upper,dest1)
lower
= pattern[i][0] - 25
upper
= pattern[i][0] + 25
cv.InRangeS(channels[
2],lower,upper,dest2)
cv.And(dest0,dest1,dest3)
temp
= cv.CreateImage(cv.GetSize(img),8,1)
cv.And(dest2,dest3,temp)
result.append(temp)

cv.Or(result[0],result[
1],dest0)
cv.Or(dest0,result[
2],dest3)
cv.NamedWindow(
"result",cv.CV_WINDOW_AUTOSIZE)
cv.ShowImage(
"result",dest3)
#cv.WaitKey(0)
return dest3

def xyProject(mat,imagesize):
"""
"project" image to x and y axis
"""
colmask
=cv.CreateMat(imagesize[1],1,cv.CV_8UC1)
rowmask
=cv.CreateMat(1,imagesize[0],cv.CV_8UC1)
cv.Set(colmask,
1)
cv.Set(rowmask,
1)
colsum
=[]

for i in range(imagesize[0]):
col
= cv.GetCol(matrix,i)
a
= cv.DotProduct(colmask,col)
colsum.append(a)

rowsum
=[]
for i in range(imagesize[1]):
row
= cv.GetRow(matrix,i)
a
= cv.DotProduct(rowmask,row)
rowsum.append(a)

return (colsum,rowsum)

def getMinNonZero(arr):
minv
= max(arr)
for a in arr:
if a > 0 and a < minv:
minv
= a
return minv

def getMeanDistance(data,minv):
shiftindex
= [0]
index
=[]

for i in range(len(data)):
if data[i]>minv:
index.append(i)
shiftindex.append(i)

index.append(index[len(index)
-1])
dist
=[]
for i in range(len(index)):
dist.append(index[i]
-shiftindex[i])

maxdist
=max(dist)
index
=[]
for i in dist:
if i != maxdist and i > 1:
index.append(i)
mdist
= sum(index)/len(index)
#print mdist
return mdist

if __name__ == "__main__":
#TODO: robost against illumination change
pattern_colors=[(219,235,102),(213,71,155),(67,170,228)]#pink, cyan, yellow
image = cv.LoadImage(names[0])
imagesize
=cv.GetSize(image)

matrix
= findFirstColorPattern(image,pattern_colors)
(colsum,rowsum)
= xyProject(matrix,imagesize)

# the following code is trying to find the edge of the small color rectangles
rx0 = 0
minv1
= getMinNonZero(colsum)
for a in colsum:
if a > minv1:
break
rx0
+= 1

minv2
= getMinNonZero(rowsum)
ry0
= 0
for a in rowsum:
if a > minv2:
break
ry0
+= 1

arrlen
= len(colsum)
rx1
= arrlen -1
for i in range(arrlen):
if colsum[arrlen-1-i] > minv1:
break
rx1
-= 1

arrlen
= len(rowsum)
ry1
= arrlen -1
for i in range(arrlen):
if rowsum[arrlen-1-i] > minv2:
break
ry1
-= 1

# try to compute the mean distance between two small color rectangles
mdist = getMeanDistance(colsum,minv1)
rectx0
= rx0-mdist*2
recty0
= ry0-mdist*2
if rectx0 < 0:
rectx0
= 0
if recty0 < 0:
recty0
= 0
rectx1
= rx1+mdist*2
if rectx1 > imagesize[0]:
rectx1
= imagesize[0]
recty1
= ry1+mdist*2
if recty1 > imagesize[1]:
recty1
= imagesize[1]

#cv.Circle(image,(int(rx0),int(ry0)),5,cv.CV_RGB(0,255,0))
#cv.Circle(image,(int(rx1),int(ry1)),5,cv.CV_RGB(0,255,0))
roi = cv.GetSubRect(image,(rectx0,recty0,rectx1-rectx0,recty1-recty0))
pt
=[(rectx0,recty0),(rectx1,recty0),(rectx1,recty1),(rectx0,recty1)]
cv.PolyLine( image, [pt],
1, cv.CV_RGB(0,255,0), 3, cv.CV_AA, 0 );

cv.NamedWindow(
"test",cv.CV_WINDOW_AUTOSIZE)
cv.ShowImage(
"test",image)
cv.NamedWindow(
"roi",cv.CV_WINDOW_AUTOSIZE)
cv.ShowImage(
"roi",roi)
cv.SaveImage(
"result.png",image)
cv.WaitKey(0)

cv.DestroyWindow(
"test")
cv.DestroyWindow(
"result")
cv.DestroyWindow(
"roi")

 

 


posted on 2010-12-23 00:05  cornsea  阅读(2529)  评论(0编辑  收藏  举报

导航