python图像分割识别_python libsvm 图像分割 字符识别 | 学步园

Python 用的是2.7版本

字符识别用的是libsvm，针对特定的数据进行训练，之后在进行识别，识别率还是比较高的，之前我也用过mnist数据集，结果不是太令人满意

#coding=utf-8

#!/usr/bin/env python

import Image

import cv2

from cv2 import cv

import numpy as np

from pylab import *

import glob

import os

from svmutil import *

'''

分割出图片中的数字，并改变大小，返回值为图片数组 ，如path = 'D:/pic/93.png'

返回的数据格式[[],[],[],.....]

'''

def picSplitResize(path):

image = cv2.imread(path)

''' 获取一张图片四个角的灰度值'''

w,h = cv.GetSize(cv.fromarray(image))

gray = cv2.cvtColor(image,cv2.COLOR_BGR2GRAY)

angleGray = [0] * 4

angleGray[0] = gray[0][0]

angleGray[1] = gray[0][w-1]

angleGray[2] = gray[h-1][0]

angleGray[3] = gray[h-1][w-1]

num = 0

for i in range(len(angleGray)):

if angleGray[i] > 100:

num += 1

#print num

''' num大于3说明四个角是白色的，否则是黑色'''

if num >=3:

#把原来颜色反转后加强

ret , bin = cv2.threshold(gray, 100,255, cv2.THRESH_BINARY_INV)

else:

#保持原来颜色不变，只是加强

ret , bin = cv2.threshold(gray, 100,255, cv2.THRESH_BINARY)

#cv2.imshow("bin",bin)

#膨胀后腐蚀

dilated = cv2.dilate(bin, cv2.getStructuringElement(cv2.MORPH_RECT,(2, 2)))

eroded = cv2.erode(dilated, cv2.getStructuringElement(cv2.MORPH_RECT,(2, 2)))

# 腐蚀后膨胀

eroded = cv2.erode(eroded, cv2.getStructuringElement(cv2.MORPH_RECT,(2, 2)))

dilated = cv2.dilate(eroded, cv2.getStructuringElement(cv2.MORPH_RECT,(2, 2)))

#细化

median = cv2.medianBlur(dilated, 3)

median1 = cv2.medianBlur(dilated, 3)

#轮廓查找,查找前必须转换成黑底白字

contours, heirs = cv2.findContours(median1,cv2.RETR_EXTERNAL,cv2.CHAIN_APPROX_SIMPLE)

i = 0

pic = []

dictPic = {}

for tours in contours:

rc = cv2.boundingRect(tours)

#rc[0] 表示图像左上角的纵坐标，rc[1] 表示图像左上角的横坐标，rc[2] 表示图像的宽度，rc[3] 表示图像的高度，

#cv2.rectangle(bin, (rc[0],rc[1]),(rc[0]+rc[2],rc[1]+rc[3]),(255,0,255))

image1M = cv.fromarray(median)

image1Ip = cv.GetImage(image1M)

cv.SetImageROI(image1Ip,rc)

imageCopy = cv.CreateImage((rc[2], rc[3]),cv2.IPL_DEPTH_8U, 1)

cv.Copy(image1Ip,imageCopy)

cv.ResetImageROI(image1Ip)

#print np.asarray(cv.GetMat(imageCopy))

#把图像左上角的纵坐标和图像的数组元素放到字典里

dictPic[rc[0]] = np.asarray(cv.GetMat(imageCopy))

pic.append(np.asarray(cv.GetMat(imageCopy)))

#cv.ShowImage(str(i), imageCopy)

#cv.Not(imageCopy, imageCopy) #函数cvNot(const CvArr* src,CvArr* dst)会将src中的每一个元素的每一位取反，然后把结果赋给dst

#cv.SaveImage(str(i)+ '.jpg',imageCopy)

i = i+1

sortedNum = sorted(dictPic.keys())

for i in range(len(sortedNum)):

pic[i] = dictPic[sortedNum[i]]

#cv2.waitKey(0)

return pic

'''

#调整图片大小，先归一化图片，并把原图片放在中间,返回的数据格式 [[],[],.....]

'''

def resize(picArray,size):

picNew = []

for i in range(len(picArray)):

imgPIL = Image.fromarray(picArray[i])

h,w = imgPIL.size

newH = w//2 - h//2 #把图片放在中间

imgEmpty = Image.new('L',(w,w),0) #创建一张背景为黑色的图片

imgEmpty.paste(imgPIL,(newH,0))

imgResize = imgEmpty.resize(size,Image.ANTIALIAS)

imgResize0255 = imgResize.point(lambda x: 255 if x > 10 else 0) #0是黑，255是白 黑白加强

imgResizeArray = array(imgResize0255).flatten().tolist() #转换为一维

imgResizeArraySmaller = [float(x)/255 for x in imgResizeArray]#把0-255转成0-1

#print array(imgResize0255)

picNew.append(imgResizeArraySmaller)

#imgResize0255.show()

imgResize0255.save('D:/pic/result/' + str(i) + '.jpg')

return picNew

'''

#获取训练数据,path为文件夹路径,suffix为图片后缀，图片名字的首字母必须是图片中的内容

'''

def traindata(path,suffix):

train_images = []

train_labels = []

for files in glob.glob(path + '/*.' + suffix):

filepath,filename = os.path.split(files)

train_labels.append(int(filename[0:1]))

pic = picSplitResize(filepath + '/' + filename)

picNew = resize(pic,(20,20))

train_images.append(picNew[0])

#im = Image.open(filepath + '/' + filename)

return train_images,train_labels

'''

#创建LibSVM分类器,返回值为识别出的内容

'''

def predictPIC(train_images,train_labels,picdata):

prob = svm_problem(train_labels,train_images)

param = svm_parameter('-t 2')

prob = svm_train(prob,param)

labels = [0] * len(picdata)

flag = svm_predict(labels,picdata,prob)

print flag[0]

pic = picSplitResize('D:/pic/12.png')

#pic = picSplitResize('D:/pic/train and test/train/9-5.bmp')

#pic = picSplitResize('D:/pic/train and test/test1/7.bmp')

picdata = resize(pic,(20,20))

#train_images,train_labels = traindata('D:/pic/train and test/train','bmp')

#result = predictPIC(train_images,train_labels,picdata)

用于测试的图片

分割的结果(只贴出来部分)

cv2.imread()读入的图像使用NumPy数组表示，而通过cv.LoadImage()读入的图像为iplimage对象

array=cv2.imread("lena.jpg")iplimage=cv.LoadImage("lena.jpg")

最后

以上就是俏皮雨为你收集整理的python图像分割识别_python libsvm 图像分割 字符识别 | 学步园的全部内容，希望文章能够帮你解决python图像分割识别_python libsvm 图像分割 字符识别 | 学步园所遇到的程序开发问题。

如果觉得靠谱客网站的内容还不错，欢迎将靠谱客网站推荐给程序员好友。