HR/LR文件夹灰度图像读取后显示偏绿问题排查及灰度显示方案咨询

问题根源 #

1. 维度不匹配导致数据异常：你设置的目标图像尺寸是3通道格式（如(256,256,3)），但通过as_gray=True读取的是单通道灰度图，两者维度不兼容。transform.resize在处理时会错误调整数据结构，后续显示时触发异常渲染。
2. 默认配色导致绿色显示：matplotlib对单通道图像默认使用viridis色系（绿色调），没有指定灰度配色方案时，就会呈现绿色。

修正步骤 #

1. 统一图像维度 #

如果仅需处理灰度图，将目标尺寸改为单通道格式：

low_resolution_shape = (64, 64)
high_resolution_shape = (256, 256)

如果后续模型需要3通道输入，可在resize后显式扩展通道（替代原注释的代码）：

# 在resize后添加通道扩展代码
hr_img1_high_resolution = np.repeat(hr_img1_high_resolution[..., np.newaxis], 3, axis=-1)
lr_img1_low_resolution = np.repeat(lr_img1_low_resolution[..., np.newaxis], 3, axis=-1)

2. 指定灰度显示配色 #

在imshow调用时添加cmap='gray'参数，强制使用灰度配色：

# 显示高分辨率图
axs[i].imshow(high_resolution_images[i], cmap='gray')

# 显示低分辨率图
axs[i].imshow(low_resolution_images[i], cmap='gray')

完整修正代码 #

from skimage import io, transform
import os
import numpy as np
import matplotlib.pyplot as plt
import glob

def sample_images(data_dir, batch_size, high_resolution_shape, low_resolution_shape):
    hr_dir = os.path.join(data_dir, 'hr')  # 高分辨率图像文件夹路径
    lr_dir = os.path.join(data_dir, 'lr')  # 低分辨率图像文件夹路径

    # 获取HR和LR目录下所有图像的路径列表
    hr_images = glob.glob(os.path.join(hr_dir, '*.*'))
    lr_images = glob.glob(os.path.join(lr_dir, '*.*'))

    # 随机选择一批图像
    hr_images_batch = np.random.choice(hr_images, size=batch_size)
    lr_images_batch = np.random.choice(lr_images, size=batch_size)

    low_resolution_images = []
    high_resolution_images = []

    for hr_img, lr_img in zip(hr_images_batch, lr_images_batch):
        # 读取当前HR图像为单通道灰度图
        hr_img1 = io.imread(hr_img, as_gray=True)

        # 调整HR图像尺寸
        hr_img1_high_resolution = transform.resize(hr_img1, high_resolution_shape, mode='constant')
        # 如果需要3通道输入，取消下面一行注释
        # hr_img1_high_resolution = np.repeat(hr_img1_high_resolution[..., np.newaxis], 3, axis=-1)

        # 读取当前LR图像为单通道灰度图
        lr_img1 = io.imread(lr_img, as_gray=True)

        # 调整LR图像尺寸
        lr_img1_low_resolution = transform.resize(lr_img1, low_resolution_shape, mode='constant')
        # 如果需要3通道输入，取消下面一行注释
        # lr_img1_low_resolution = np.repeat(lr_img1_low_resolution[..., np.newaxis], 3, axis=-1)

        # 数据增强：随机水平翻转
        if np.random.random() < 0.5:
            hr_img1_high_resolution = np.fliplr(hr_img1_high_resolution)
            lr_img1_low_resolution = np.fliplr(lr_img1_low_resolution)

        high_resolution_images.append(hr_img1_high_resolution)
        low_resolution_images.append(lr_img1_low_resolution)
    
    # 将列表转换为numpy数组并返回
    return np.array(high_resolution_images), np.array(low_resolution_images)


data_dir = r'C:\\Users\\hp\\Desktop\\dataasets\\Newfolder\\data\\'
batch_size = 2
low_resolution_shape = (64, 64)  # 单通道灰度图尺寸
high_resolution_shape = (256, 256)  # 单通道灰度图尺寸

# 采样一批图像
high_resolution_images, low_resolution_images = sample_images(data_dir=data_dir, batch_size=batch_size,
                                                              low_resolution_shape=low_resolution_shape,
                                                              high_resolution_shape=high_resolution_shape)

# 显示高分辨率图像
fig, axs = plt.subplots(nrows=1, ncols=batch_size, figsize=(12, 4))
for i in range(batch_size):
    axs[i].imshow(high_resolution_images[i], cmap='gray')
    axs[i].axis('off')
plt.suptitle('高分辨率图像')
plt.show()

# 显示低分辨率图像
fig, axs = plt.subplots(nrows=1, ncols=batch_size, figsize=(12, 4))
for i in range(batch_size):
    axs[i].imshow(low_resolution_images[i], cmap='gray')
    axs[i].axis('off')
plt.suptitle('低分辨率图像')
plt.show()

内容的提问来源于stack exchange，提问作者stella