import tensorflow as tf
import glob, os,io,sys
import pandas as pd
import numpy as np
os.environ["TF_CPP_MIN_LOG_LEVEL"]='2'

def read_pic():
    file_list=glob.glob("./code_imgs/*.png")
    file_queue=tf.train.string_input_producer(file_list)
    reader=tf.WholeFileReader()
    filename,image=reader.read(file_queue)
    decoded=tf.image.decode_png(image,channels=3)
    decoded.set_shape([28,96,3])
    image_cast=tf.cast(decoded,tf.float32)
    filename_batch,image_batch=tf.train.batch([filename,image_cast],batch_size=40,num_threads=2,capacity=40)
    return filename_batch,image_batch

def parse_csv():
    csv_data=pd.read_csv('labels.csv',names=['file_num','chars'],index_col='file_num')
    labels=[]
    for label in csv_data["chars"]:
        letter=[]
        for word in label:
            letter.append(ord(word)-ord('a'))
        labels.append(letter)
    csv_data['labels']=labels
    return csv_data

def filename2label(filenames,csv_data):
    labels=[]
    for filename in filenames:
        file_num="".join(filter(str.isdigit,str(filename)))
        target=csv_data.loc[int(file_num),"labels"]
        labels.append(target)
    return np.array(labels)

def create_weights(shape):
    return tf.Variable(initial_value=tf.random_normal(shape=shape,stddev=0.01))

def create_model(x):
    with tf.variable_scope('conv1'):
        conv1_weights=create_weights(shape=[5,5,3,32])
        conv1_bias=create_weights(shape=[32])
        conv1_x=tf.nn.conv2d(input=x,filter=conv1_weights,strides=[1,1,1,1],padding='SAME')+conv1_bias
        relu1_x=tf.nn.relu(conv1_x)
        pool1_x=tf.nn.max_pool(value=relu1_x,ksize=[1,2,2,1],strides=[1,2,2,1],padding='SAME')
   
    with tf.variable_scope('conv2'):
        conv2_weights=create_weights(shape=[5,5,32,64])
        conv2_bias=create_weights(shape=[64])
        conv2_x=tf.nn.conv2d(input=pool1_x,filter=conv2_weights,strides=[1,1,1,1],padding='SAME')+conv2_bias
        relu2_x=tf.nn.relu(conv2_x)
        pool2_x=tf.nn.max_pool(value=relu2_x,ksize=[1,2,2,1],strides=[1,2,2,1],padding='SAME')
    
    with tf.variable_scope('full_connection'):
        x_fc=tf.reshape(pool2_x,shape=[-1,7*24*64])
        weights_fc=create_weights(shape=[7*24*64,104])
        bias_fc=create_weights(shape=[104])
        y_predict=tf.matmul(x_fc,weights_fc)+bias_fc

    return y_predict

def list2text(textlist):
    tm=[]
    for i in textlist:
        tm.append(chr(97+i))
    return "".join(tm)

def train():
    filename,image=read_pic()
    csv_data=parse_csv()
    x=tf.placeholder(tf.float32,shape=[None,28,96,3])
    y_true=tf.placeholder(tf.float32,shape=[None,104])
    y_predict=create_model(x)
    loss_list=tf.nn.sigmoid_cross_entropy_with_logits(labels=y_true,logits=y_predict)
    loss=tf.reduce_mean(loss_list)
    optimizer=tf.train.AdamOptimizer(learning_rate=0.001).minimize(loss)#優化損失
    equal_list=tf.reduce_all(tf.equal(tf.argmax(tf.reshape(y_predict,shape=[-1,4,26]),axis=2), tf.argmax(tf.reshape(y_true,shape=[-1,4,26]),axis=2)),axis=1)
    accurary=tf.reduce_mean(tf.cast(equal_list,tf.float32))
    init=tf.global_variables_initializer()
    saver=tf.train.Saver()
    with tf.Session() as sess:
        sess.run(init)
        coord=tf.train.Coordinator()
        threads=tf.train.start_queue_runners(sess=sess,coord=coord)
        try:
            for i in range(10000):
                filename_val,image_val=sess.run([filename,image])
                labels=filename2label(filename_val,csv_data)
                labels_value=tf.reshape(tf.one_hot(labels,depth=26),[-1,104]).eval()
                _,error,accurary_value=sess.run([optimizer,loss,accurary],feed_dict={x:image_val,y_true:labels_value})
                print("The %d Train Result---loss:%f,accurary:%f" % (i+1,error,accurary_value))
                if accurary_value>0.99:
                    saver.save(sess,'model/crack_captcha.model99',global_step=i)
                    break
        except tf.errors.OutOfRangeError:
            print("done ,now let's kill all threads")
        finally:
            coord.request_stop()
            print("all threads ask stop")
        coord.join(threads)
        print("all thread stopped")

def crackcaptcha():
    truetext=[]
    with open('a.txt','r') as f:
        for filename in f.readlines():
            truetext.append(filename.strip('\r\n'))
    dis='False'
    goodnum=0

    x=tf.placeholder(tf.float32,shape=[None,28,96,3])
    y_true=tf.placeholder(tf.float32,shape=[None,104])
    keep_prob = tf.placeholder(tf.float32)
    y_predict=create_model(x)
    saver=tf.train.Saver()
    with tf.Session() as sess:
        sess.run(tf.global_variables_initializer())
        checkpoint=tf.train.get_checkpoint_state('model')
        if checkpoint and checkpoint.model_checkpoint_path:
            saver.restore(sess,checkpoint.model_checkpoint_path)
            print("successfully loaded:",checkpoint.model_checkpoint_path)
        else:
            print("Could not found model files")
        
        for i in range(1,201):
            image=tf.read_file('crackimgs/'+str(i)+'.png')
            decoded=tf.image.decode_png(image,channels=3)
            decoded.set_shape([28,96,3])
            decoded_val=sess.run(decoded)
            image=np.array(decoded_val)

            predict=tf.argmax(tf.reshape(y_predict,[-1,4,26]),2)
            outtext=sess.run(predict,feed_dict={x:[image],keep_prob:1})
            cracktext=list2text(outtext[0].tolist())
            if cracktext==truetext[i-1]:
                goodnum+=1
                dis='True'
            else:
                dis='False'
            print('The {} Image Content is:{},Your Crack Word is :{},Result:{}'.format(i,truetext[i-1],cracktext,dis))
        print('The End accurary is:{}%'.format(goodnum/200*100))

if __name__=='__main__':
    train()
    crackcaptcha()