tensorflow使用Image结果可视化，图像结果出现乱码问题

from tensorflow.examples.tutorials.mnist import input_data
import tensorflow as tf
import conv_set as c
import read_data as r
from PIL import Image
import conv_set as c
import numpy as np
import matplotlib.pyplot as plt

TRAIN_DATA_PATH = './train.tfrecords'
TEST_DATA_PATH = './train.tfrecords'

LEARNING_RATE = 1e-5
BATCH_SIZE = 1
BATCH_SIZE_TEST = 10000
MODEL_DIR = './model/'
NUM_EPOCHS = 10

xs = tf.placeholder(tf.float32, [1, 28, 28, 1])  # 输入图像
ys = tf.placeholder(tf.float32, [None, 10])  # 输入分类结果

with tf.variable_scope('reshape'):
    x_image = tf.reshape(xs, [28, 28])  # 28*28*1

    # 迭代器，读数据
with tf.variable_scope('input_data'):
    train_dataset = r.read_and_decode(TRAIN_DATA_PATH, BATCH_SIZE, NUM_EPOCHS)  # 读数据
    train_iterator = train_dataset.make_initializable_iterator()
    sess = tf.InteractiveSession()
    sess.run(train_iterator.initializer)
    img_batch, label_batch = train_iterator.get_next()
    sess.run(tf.global_variables_initializer())


def method1(img, sess):
    x_result2 = sess.run(x_image, feed_dict={xs: img})
    return x_result2


def method2(img, sess):
    x_rsp = tf.reshape(img, [28, 28])
    x_result = sess.run(x_rsp)
    return x_result


with tf.variable_scope('writer'):
    for i in range(100):
        img, label = sess.run([img_batch, label_batch])
        x_result = method1(img, sess)
        I = Image.fromarray(x_result,'L')
        I.save('./Save_test' + '/' + str(i) + '_''Label_''.jpg')  # 存下图片



        # plt.title('conv1')
        # plt.imshow(I)
        # plt.show()
    # writer = tf.summary.FileWriter('./log', sess.graph)  # 保存tensorboard
    # writer.add_summary(img_scalar)

Method1的结果

Method2的结果
原因：
I = Image.fromarray(x_result,‘L’)
改为：
I = Image.fromarray(x_result)
但是这样保存会报错，无法识别是jpg格式，只能用plt方法show出来

或者：
xs = tf.placeholder(tf.float32, [1, 28, 28, 1]) # 输入图像
改为
xs = tf.placeholder(tf.uint8, [1, 28, 28, 1]) # 输入图像 （因为保存tfrecord用uint8）