试卷批改系统.py

import numpy as np
import tensorflow as tf
import cv2
from PIL import Image
image = cv2.imread('D:\opencv_camera\digits\\12.jpg')
lower=np.array([50,50,150])
upper=np.array([140,140,250])
mask = cv2.inRange(image, lower, upper)
output1 = cv2.bitwise_and(image, image, mask=mask)

output=cv2.cvtColor(output1,cv2.COLOR_BGR2GRAY)
th2 = cv2.adaptiveThreshold(output,255,cv2.ADAPTIVE_THRESH_GAUSSIAN_C,cv2.THRESH_BINARY,11,0)
# cv2.imshow("images",th2)
# cv2.waitKey(0)
roi=[]
for i in range(8):
    img_roi = output1[260:310,35+55*i:80+58*i]
    roi.append(img_roi)
#用来分割图片
def picture(x):
    shuzu = []
    left=0
    right=x.shape[1]
    for i in range(x.shape[1]):
        if sum(map(sum,x[1:50,i:i+1]))>0:
            left=i
            break
    for j in range(left,x.shape[1]):
        if sum(map(sum,x[1:50,j:j+1]))==0:
            right=j+2
            break
    leftimg=x[1:50,left:right]
    shuzu.append(leftimg)
    rightimg=x[1:50,right:x.shape[1]-5]
    shuzu.append(rightimg)
    return shuzu
def normalizepic(pic):
    im_arr = pic
    im_nparr = []
    for x in im_arr:
        x=1-x/255
        im_nparr.append(x)
    im_nparr = np.array([im_nparr])
    return im_nparr
# cv2.imshow("images",roi[0])
# cv2.waitKey(0)
img1=[]
for i in range(8):
    output=cv2.cvtColor(roi[i],cv2.COLOR_BGR2GRAY)
    th2 = cv2.adaptiveThreshold(output,255,cv2.ADAPTIVE_THRESH_GAUSSIAN_C,cv2.THRESH_BINARY,11,0)
    # cv2.imshow("images", th2)
    # cv2.waitKey(0)
    img_seq = picture(th2)
    for j in range(len(img_seq)):
        #cv2.imshow("images",img_seq[j])
        # cv2.waitKey(0)
        ret, thresh2 = cv2.threshold(img_seq[j], 0, 255, cv2.THRESH_BINARY_INV)
        cv2.imshow("images", thresh2)
        cv2.waitKey(0)
        res = cv2.resize(thresh2, (28, 28), interpolation=cv2.INTER_CUBIC)

        img=normalizepic(res).reshape((1,784))
        img = img.astype(np.float32)
        img1.append(img)

########图片预处理##########

def weight_variable(shape):
    initial=tf.truncated_normal(shape,stddev=0.1)
    return tf.Variable(initial,dtype=tf.float32,name='weight')
def bias_variable(shape):
    initial=tf.constant(0.1,shape=shape)
    return tf.Variable(initial,dtype=tf.float32,name='biases')
def conv2d(x,W):
    return tf.nn.conv2d(x,W,strides=[1,1,1,1],padding='SAME')
def max_pool_2x2(x):
    return tf.nn.max_pool(x,ksize=[1,2,2,1],strides=[1,2,2,1],padding='SAME')

xs=tf.placeholder(tf.float32,[None,784])#输入是一个28*28的像素点的数据

keep_prob=tf.placeholder(tf.float32)

x_image=tf.reshape(xs,[-1,28,28,1])#xs的维度暂时不管，用-1表示，28,28表示xs的数据，1表示该数据是一个黑白照片，如果是彩色的，则写成3
#卷积层1
W_conv1=weight_variable([5,5,1,32])#抽取一个5*5像素，高度是32的点,每次抽出原图像的5*5的像素点，高度从1变成32
b_conv1=bias_variable([32])
h_conv1=tf.nn.relu(conv2d(x_image,W_conv1)+b_conv1)#输出 28*28*32的图像
h_pool1=max_pool_2x2(h_conv1)##输出14*14*32的图像，因为这个函数的步长是2*2，图像缩小一半。
#卷积层2
W_conv2=weight_variable([5,5,32,64])#随机生成一个5*5像素，高度是64的点,抽出原图像的5*5的像素点，高度从32变成64
b_conv2=bias_variable([64])
h_conv2=tf.nn.relu(conv2d(h_pool1,W_conv2)+b_conv2)#输出14*14*64的图像
h_pool2=max_pool_2x2(h_conv2)##输出7*7*64的图像，因为这个函数的步长是2*2，图像缩小一半。
#fully connected 1
W_fc1=weight_variable([7*7*64,1024])
b_fc1=bias_variable([1024])
h_pool2_flat=tf.reshape(h_pool2,[-1,7*7*64])#将输出的h_pool2的三维数据变成一维数据，平铺下来，（-1）代表的是有多少个例子
h_fc1=tf.nn.relu(tf.matmul(h_pool2_flat,W_fc1)+b_fc1)
h_fc1_drop=tf.nn.dropout(h_fc1,keep_prob)
#fully connected 2
W_fc2=weight_variable([1024,10])
b_fc2=bias_variable([10])
prediction=tf.nn.softmax(tf.matmul(h_fc1_drop,W_fc2)+b_fc2)#输出层

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    saver = tf.train.Saver()
    saver.restore(sess, "my_net/save_net.ckpt")
    result= tf.argmax(prediction, 1)
    #result= result.eval(feed_dict={xs:img, keep_prob: 1.0}, session=sess)
    for i in range(16):
        result=sess.run(prediction, feed_dict={xs:img1[i],keep_prob: 1.0})
        result1 =np.argmax(result,1)
        print(result1)