TensorFlow中的图像处理

た入场券 2023-06-27 04:27 1阅读 0赞

本文具体数据集与源代码可从我的GitHub地址获取  
[https://github.com/liuzuoping/Deep\_Learning\_note][https_github.com_liuzuoping_Deep_Learning_note]

## 数据预处理 ##

### 编码处理 ###

*  以下代码示范对 JPEG 格式图像的解码处理

#--coding--:utf-8
    import tensorflow as tf
    import matplotlib.pyplot as plt
    
    # 读取图像的原始数据
    image_raw_data = tf.gfile.FastGFile('images/plate1.jpg', 'rb').read()
    
    
    # 使用pyplot显示图像
    def show(img_data):
        plt.imshow(img_data.eval())
        plt.show()
    
    
    with tf.Session() as sess:
        # 将原始数据解码成多维矩阵
        img_data = tf.image.decode_jpeg(image_raw_data)
        print(img_data.eval())
        show(img_data)
    
        # 将图像的矩阵编码成图像并存入文件
        encoded_image = tf.image.encode_jpeg(img_data)
        with tf.gfile.GFile('preprocess/output.jpg', 'wb') as f:
            f.write(encoded_image.eval())
    
        # 将图像数据的类型转为实数类型，便于对图像进行处理
        img_data = tf.image.convert_image_dtype(img_data, dtype=tf.float32)

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpdXp1b3Bpbmc_size_16_color_FFFFFF_t_70]

### 大小调整 ###

*  以下代码都是在图像编码处理代码的基础下运行，省去了加载原始图像，定义会话等过程

# 用resize_images调整图像大小
    # 第一个参数为原始图像
    # 第二个参数为调整后的图像大小[new_height,new_width]，跟旧版本分为两个参数不一样
    # method参数给出了调整图像大小的算法
    with tf.Session() as sess:
        # 将原始数据解码成多维矩阵
        img_data = tf.image.decode_jpeg(image_raw_data)
        #print(img_data.eval())
        print ("原始图像:")
        show(img_data)
    
        # 将图像的矩阵编码成图像并存入文件
        encoded_image = tf.image.encode_jpeg(img_data)
        with tf.gfile.GFile('preprocess/output.jpg', 'wb') as f:
            f.write(encoded_image.eval())
    
        # 将图像数据的类型转为实数类型，便于对图像进行处理
        img_data = tf.image.convert_image_dtype(img_data, dtype=tf.float32)
        resized = tf.image.resize_images(img_data, [300, 300], method=0)
        #print(resized.get_shape()) # 图像深度没有显式指定则为问号
        print ("裁剪后的图像:")
        show(resized)
        
        # 用resize_image_with_crop_or_pad调整图像大小
        # 第一个参数为原始图像
        # 第二个和第三个参数是调整后的图像大小，大于原图则填充，小于则裁剪居中部分
        croped = tf.image.resize_image_with_crop_or_pad(img_data, 200, 200)
        print ("通过resize_image_with_crop_or_pad处理后的图像，此图为裁剪后:")
        show(croped)
        padded = tf.image.resize_image_with_crop_or_pad(img_data, 500, 700)
        print ("通过resize_image_with_crop_or_pad处理后的图像，此图为填充后:")
        show(padded)
        
        # 用central_crop调整图像大小
        # 第一个参数是原始图像
        # 第二个参数为调整比例，是(0,1]的实数
        central_cropped = tf.image.central_crop(img_data, 0.5)
        print ("通过比例裁剪后的图像:")
        show(central_cropped)

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpdXp1b3Bpbmc_size_16_color_FFFFFF_t_70 1]  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpdXp1b3Bpbmc_size_16_color_FFFFFF_t_70 2]  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpdXp1b3Bpbmc_size_16_color_FFFFFF_t_70 3]  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpdXp1b3Bpbmc_size_16_color_FFFFFF_t_70 4]  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpdXp1b3Bpbmc_size_16_color_FFFFFF_t_70 5]

### 图像翻转 ###

*  以下代码示范了将图像进行上下翻转、左右翻转及沿对角线翻转

with tf.Session() as sess:
        # 将原始数据解码成多维矩阵
        img_data = tf.image.decode_jpeg(image_raw_data)
        #print(img_data.eval())
        print ("原始图像:")
        show(img_data)
        
        # 图像翻转
        flipped = tf.image.flip_up_down(img_data)  # 上下
        print ("上下翻转后:")
        show(flipped)
        flipped = tf.image.flip_left_right(img_data)  # 左右
        print ("左右翻转后:")
        show(flipped)
        transposed = tf.image.transpose_image(img_data)  # 对角线
        print ("沿对角线反转后: ")
        show(transposed)
        
        # 随机翻转图像
        flipped = tf.image.random_flip_up_down(img_data)  # 随机上下
        print ("随机上下翻转后:")
        show(flipped)
        flipped = tf.image.random_flip_left_right(img_data)  # 随机左右
        print ("随机左右翻转后:")
        show(flipped)

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpdXp1b3Bpbmc_size_16_color_FFFFFF_t_70 6]  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpdXp1b3Bpbmc_size_16_color_FFFFFF_t_70 7]  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpdXp1b3Bpbmc_size_16_color_FFFFFF_t_70 8]

### 色彩调整 ###

图像的色彩调整有四方面：

*  亮度
 *  对比度
 *  饱和度
 *  色相

with tf.Session() as sess:
        # 将原始数据解码成多维矩阵
        img_data = tf.image.decode_jpeg(image_raw_data)
        #print(img_data.eval())
        print ("原始图像:")
        show(img_data)
    
        # 调整图像的亮度
        adjusted = tf.image.adjust_brightness(img_data, 0.5)  # 将图像的亮度+0.5
        print ("将图像的亮度+0.5:")
        show(adjusted)
        adjusted = tf.image.random_brightness(
            img_data, max_delta=0.5)  # 在[-0.5,0.5]范围内随机调整图像亮度
        print ("在[-0.5,0.5]范围内随机调整图像亮度:")
        show(adjusted)
        
        # 调整图像的对比度
        adjusted = tf.image.adjust_contrast(img_data, -5)  # 将图像的对比度-5
        print ("将图像的对比度-5:")
        show(adjusted)
        adjusted = tf.image.adjust_contrast(img_data, 5)  # 将图像的对比度+5
        print ("将图像的对比度+5:")
        show(adjusted)
        adjusted = tf.image.random_contrast(img_data, lower=1, upper=5)  # 随机调整对比度
        
        # 调整图像的饱和度
        adjusted = tf.image.adjust_saturation(img_data, -5)  # 将饱和度-5
        print ("将饱和度-5:")
        show(adjusted)
        adjusted = tf.image.adjust_saturation(img_data, 5)  # 将饱和度+5
        print ("将饱和度+5:")
        show(adjusted)
        adjusted = tf.image.random_saturation(img_data, lower=1, upper=5)  # 随机调整饱和度
        
        # 调整图像的色相
        adjusted = tf.image.adjust_hue(img_data, 0.5)  # 将色相+0.5
        print ("将色相+0.5:")
        show(adjusted)
        adjusted = tf.image.random_hue(img_data, max_delta=0.5)  # 随机调整色相
        print ("随机调整色相:")
        show(adjusted)

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpdXp1b3Bpbmc_size_16_color_FFFFFF_t_70 9]  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpdXp1b3Bpbmc_size_16_color_FFFFFF_t_70 10]  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpdXp1b3Bpbmc_size_16_color_FFFFFF_t_70 11]  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpdXp1b3Bpbmc_size_16_color_FFFFFF_t_70 12]  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpdXp1b3Bpbmc_size_16_color_FFFFFF_t_70 13]

[https_github.com_liuzuoping_Deep_Learning_note]: https://github.com/liuzuoping/Deep_Learning_note
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpdXp1b3Bpbmc_size_16_color_FFFFFF_t_70]: https://img-blog.csdnimg.cn/20200103135032222.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpdXp1b3Bpbmc=,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpdXp1b3Bpbmc_size_16_color_FFFFFF_t_70 1]: https://img-blog.csdnimg.cn/20200103135442230.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpdXp1b3Bpbmc=,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpdXp1b3Bpbmc_size_16_color_FFFFFF_t_70 2]: https://img-blog.csdnimg.cn/20200103135505880.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpdXp1b3Bpbmc=,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpdXp1b3Bpbmc_size_16_color_FFFFFF_t_70 3]: https://img-blog.csdnimg.cn/20200103135522662.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpdXp1b3Bpbmc=,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpdXp1b3Bpbmc_size_16_color_FFFFFF_t_70 4]: https://img-blog.csdnimg.cn/20200103135554442.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpdXp1b3Bpbmc=,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpdXp1b3Bpbmc_size_16_color_FFFFFF_t_70 5]: https://img-blog.csdnimg.cn/2020010313561182.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpdXp1b3Bpbmc=,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpdXp1b3Bpbmc_size_16_color_FFFFFF_t_70 6]: https://img-blog.csdnimg.cn/20200103135913142.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpdXp1b3Bpbmc=,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpdXp1b3Bpbmc_size_16_color_FFFFFF_t_70 7]: https://img-blog.csdnimg.cn/20200103135937237.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpdXp1b3Bpbmc=,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpdXp1b3Bpbmc_size_16_color_FFFFFF_t_70 8]: https://img-blog.csdnimg.cn/20200103135954196.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpdXp1b3Bpbmc=,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpdXp1b3Bpbmc_size_16_color_FFFFFF_t_70 9]: https://img-blog.csdnimg.cn/20200103140357938.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpdXp1b3Bpbmc=,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpdXp1b3Bpbmc_size_16_color_FFFFFF_t_70 10]: https://img-blog.csdnimg.cn/20200103140417669.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpdXp1b3Bpbmc=,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpdXp1b3Bpbmc_size_16_color_FFFFFF_t_70 11]: https://img-blog.csdnimg.cn/20200103140444155.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpdXp1b3Bpbmc=,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpdXp1b3Bpbmc_size_16_color_FFFFFF_t_70 12]: https://img-blog.csdnimg.cn/20200103140501251.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpdXp1b3Bpbmc=,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpdXp1b3Bpbmc_size_16_color_FFFFFF_t_70 13]: https://img-blog.csdnimg.cn/20200103140519244.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L2xpdXp1b3Bpbmc=,size_16,color_FFFFFF,t_70