Python实现：如何将绘制的ROI蒙版保存为无背景PNG

我用Python做图像处理，已经实现了ROI绘制功能，也写了保存ROI的代码，但当前生成的roi_region和原图尺寸完全一致，只想单独保存ROI本身（不带背景区域）。试过添加alpha通道、改用plt.figsave都没解决问题，相关代码如下：

if index == 2:
    fig_all_roi, ax_all_roi = plt.subplots()
    ax_all_roi.set_title('Extracted ROIs')

    for i, (verts_list, color) in enumerate(zip(roi_verts_list, roi_colors)):
        for verts in verts_list:
            # Create a Path object from the vertices
            path = Path(verts)
            
            # Create a mask for the region inside the ROI
            x, y = np.meshgrid(np.arange(image.shape[1]), np.arange(image.shape[0]))
            points = np.column_stack((x.flatten(), y.flatten()))
            mask = path.contains_points(points).reshape(image.shape[:2])

            # Extract the region inside the ROI from the original image
            roi_region = np.zeros_like(image, dtype=np.uint8)
            for channel in range(image.shape[2]):
                roi_region[:, :, channel] = (image[:, :, channel] * mask).astype(np.uint8)
            roi_region[:,:, 3] = (mask*255).astype(np.uint8)

            # Save the extracted ROI as an image
            save_roi_image(roi_region, i)

            # Display the extracted region in the new figure
            ax_all_roi.imshow(roi_region, extent=[0, image.shape[1], 0, image.shape[0]])

核心思路是先裁剪出ROI的最小包围矩形，再结合alpha通道实现背景透明，具体步骤如下：

1. 定位ROI的边界范围 #

根据生成的mask，找到ROI区域的最小/最大坐标，确定裁剪范围：

# 找到mask中非零区域的坐标
y_coords, x_coords = np.where(mask)
# 处理空ROI的特殊情况
if len(y_coords) == 0:
    continue
# 确定ROI的上下左右边界
x_min, x_max = x_coords.min(), x_coords.max()
y_min, y_max = y_coords.min(), y_coords.max()

2. 裁剪并生成带透明通道的ROI #

把原图和mask都裁剪到上述边界内，再拼接alpha通道：

# 裁剪原图的ROI区域
cropped_image = image[y_min:y_max+1, x_min:x_max+1]
# 裁剪对应的mask
cropped_mask = mask[y_min:y_max+1, x_min:x_max+1]

# 生成带alpha通道的RGBA图像（仅保留ROI区域，背景透明）
cropped_roi = np.dstack((cropped_image, cropped_mask.astype(np.uint8)*255))

3. 调整保存逻辑 #

保存时要选择支持透明通道的格式（比如PNG），示例修改save_roi_image函数：

from PIL import Image

def save_roi_image(roi_array, index):
    # 将numpy数组转换为PIL图像对象
    img = Image.fromarray(roi_array)
    # 保存为PNG格式（JPG不支持透明通道）
    img.save(f"roi_{index}.png")

完整修改后的代码片段 #

if index == 2:
    fig_all_roi, ax_all_roi = plt.subplots()
    ax_all_roi.set_title('Extracted ROIs')

    for i, (verts_list, color) in enumerate(zip(roi_verts_list, roi_colors)):
        for verts in verts_list:
            # Create a Path object from the vertices
            path = Path(verts)
            
            # Create a mask for the region inside the ROI
            x, y = np.meshgrid(np.arange(image.shape[1]), np.arange(image.shape[0]))
            points = np.column_stack((x.flatten(), y.flatten()))
            mask = path.contains_points(points).reshape(image.shape[:2])

            # 新增：定位ROI边界
            y_coords, x_coords = np.where(mask)
            if len(y_coords) == 0:
                continue
            x_min, x_max = x_coords.min(), x_coords.max()
            y_min, y_max = y_coords.min(), y_coords.max()

            # 裁剪ROI区域
            cropped_image = image[y_min:y_max+1, x_min:x_max+1]
            cropped_mask = mask[y_min:y_max+1, x_min:x_max+1]

            # 生成带透明通道的ROI
            cropped_roi = np.dstack((cropped_image, cropped_mask.astype(np.uint8)*255))

            # Save the extracted ROI as an image
            save_roi_image(cropped_roi, i)

            # Display the extracted region in the new figure
            ax_all_roi.imshow(cropped_roi)

内容的提问来源于stack exchange，提问作者MoBedick