当两张图片大小不同时，可以直接将较小的一张图片嵌入到较大的一张图片中。以下是具体的步骤和说明： ### 实现过程 1. **读取图像**：分别加载大图和小图。 2. **确定位置**：选择在大图上放置小图的位置 `(x,y)` ，即左上角坐标点。 3. **区域裁切**：从大图中取出与小图同样大小的一个矩形区域作为ROI（Region of Interest），其起始点就是上述选定的`(x,y)`，宽高则等于小图的尺寸。 4. **合并像素数据**：把小图的内容复制到大图对应的 ROI 区域内，这一操作实际上是对相应位置处每个颜色通道进行逐元素赋值替换。 下面提供一段Python代码示例帮助理解整个流程: ```python import cv2 # Step 1: Load two images. large_img = cv2.imread('big_picture.jpg') small_img = cv2.imread('small_logo.png') # Check that both images loaded correctly. if large_img is None or small_img is None: print("Could not open one or more files.") else: # Get dimensions for resizing if needed and positioning. rows,cols,channnels= small_img.shape # Specify where you want to place the smaller image on top left corner coords(x,y). x_offset=y_offset=50 # Ensure enough space in destination exists before proceeding further checks omitted here... # Create a Region Of Interest (ROI) within larger canvas which matches size&posn of logo . roi = large_img[y_offset:y_offset+rows , x_offset:x_offset + cols ] # Now create mask from grayscale version of watermark plus inverse also known as "background" gray=cv2.cvtColor(small_img,cv2.COLOR_BGR2GRAY) ret,binary_mask=cv2.threshold(gray,10,255,cv2.THRESH_BINARY) #threshold creates binary mask based off intensity values above certain limit e.g., black-white conversion background=np.ones_like(small_img,dtype='uint8')*255 #white colored square having same shape like original watermarked pic foreground=binary_mask #black areas correspond exactly overlying object whereas white represents transparent parts around it. # Apply some blending techniques such as AddWeighted etc.. But simplest way shown next line directly replaces pixels inside Roi with those coming only via Foreground Mask applied onto Watermark Picture itself. combined_part=cv2.bitwise_and(foreground,foreground)+cv2.bitwise_and(background^binary_mask,roi) # Finally paste modified region back into main scene picture replacing older area covered by new added content. large_img[y_offset:y_offset+rows,x_offset:x_offset+cols]=combined_part cv2.imshow('Embedded Image',large_img) cv2.waitKey() ``` 以上程序演示了如何把一个小标志图形插入进风景照的过程。需要注意的是，在真实场景里可能还需要考虑遮罩生成算法以及其它复杂情况下的处理机制等额外因素。