如何在OpenCV/Numpy中将小RGBA图像非透明像素叠加到大图中间

需求：将尺寸为(150,150,4)的RGBA格式small图像，叠加到(300,300,4)的big图像正中间。保留small中alpha通道为255的非透明像素，透明区域直接使用big对应位置的像素。现有代码实现居中后，掩码赋值阶段出现形状不匹配报错。

原错误代码 #

BACKGROUND_SIZE = 300, 300
LOGO_SIZE = 150, 150

def get_offset(bigger_shape, smaller_shape):
    logo_x_offset = int((bigger_shape[0] - smaller_shape[0]) / 2)
    logo_y_offset = int((bigger_shape[1] - smaller_shape[1]) / 2)
    return (
        logo_y_offset,
        logo_y_offset + smaller_shape[1],
        logo_x_offset,
        logo_x_offset + smaller_shape[0],
    )


small = cv2.imread(f"./small.png", flags=cv2.IMREAD_UNCHANGED)
big =  cv2.imread(f"./big.png", flags=cv2.IMREAD_UNCHANGED)
offset = get_offset(BACKGROUND_SIZE, LOGO_SIZE)
y0, y1, x0, x1 = offset

mask = np.where(small[:, :, 3] != 255)
small[mask] = big[y0:y1, x0:x1]
big[y0:y1, x0:x1] = small

报错信息 #

ValueError: shape mismatch: value array of shape (150,150,4) could not be broadcast to indexing result of shape (14899,4)

错误原因 #

np.where(small[:, :, 3] != 255)返回的是两个一维索引数组（行、列坐标），此时small[mask]取到的是所有透明像素的通道数据，形状为(N,4)（N是透明像素数量）。而big[y0:y1, x0:x1]是完整的150x150x4子图，维度不匹配导致无法广播赋值。

修复方案 #

使用布尔掩码直接匹配对应位置的像素，确保赋值双方形状一致：

• 先提取big中与small对应的子区域
• 生成布尔类型的掩码矩阵（标记small的透明区域）
• 用big子区域的对应像素替换small的透明区域
• 将处理后的small放回big的对应位置

修复后完整代码 #

import cv2
import numpy as np

BACKGROUND_SIZE = 300, 300
LOGO_SIZE = 150, 150

def get_offset(bigger_shape, smaller_shape):
    logo_x_offset = int((bigger_shape[0] - smaller_shape[0]) / 2)
    logo_y_offset = int((bigger_shape[1] - smaller_shape[1]) / 2)
    return (
        logo_y_offset,
        logo_y_offset + smaller_shape[1],
        logo_x_offset,
        logo_x_offset + smaller_shape[0],
    )


small = cv2.imread("./small.png", flags=cv2.IMREAD_UNCHANGED)
big = cv2.imread("./big.png", flags=cv2.IMREAD_UNCHANGED)
offset = get_offset(BACKGROUND_SIZE, LOGO_SIZE)
y0, y1, x0, x1 = offset

# 提取big中对应small位置的子图
big_sub = big[y0:y1, x0:x1]
# 生成布尔掩码：标记small中alpha不等于255的透明区域
mask = small[:, :, 3] != 255
# 用big子图的对应像素替换small的透明区域
small[mask] = big_sub[mask]
# 将处理后的small放回big的对应位置
big[y0:y1, x0:x1] = small

# 可选：保存结果
# cv2.imwrite("result.png", big)

内容的提问来源于stack exchange，提问作者Gameplay