OpenCV：让程序看懂图像

import cv2

img = cv2.imread("photo.jpg")                 # 读图 → NumPy 数组
print(img.shape)                              # (高, 宽, 3)，BGR 三通道

gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)  # 转 8 位灰度图
edges = cv2.Canny(gray, 100, 200)             # Canny 边缘检测

cv2.imwrite("photo_edges.png", edges)        # 按扩展名自动选编码器
cv2.imshow("edges", edges)                   # 弹窗显示（仅本地桌面）
cv2.waitKey(0)                                # 等任意按键，0 = 一直等
cv2.destroyAllWindows()

import cv2

img = cv2.imread("photo.jpg", cv2.IMREAD_GRAYSCALE)   # 直接按灰度读
edges = cv2.Canny(img, 100, 200)
print("边缘像素数:", int((edges > 0).sum()))            # 用数字确认处理有效
cv2.imwrite("photo_edges.png", edges)

h, w = img.shape[:2]
roi = img[100:300, 200:500]          # 裁剪：先行（y）后列（x）
b, g, r = cv2.split(img)             # 拆通道；cv2.merge([b, g, r]) 合回
img[50, 80]                          # 第 50 行第 80 列那个像素，返回 [B, G, R]
img[:] = 0                           # 整图涂黑——数组是引用，改了就是真改

small = cv2.resize(img, None, fx=0.5, fy=0.5)            # 缩到一半
small = cv2.resize(img, (640, 480), interpolation=cv2.INTER_AREA)

cv2.rectangle(img, (60, 50), (220, 200), (0, 255, 0), 2) # 绿框，线宽 2；-1 为实心
cv2.circle(img, (150, 120), 30, (0, 0, 255), -1)         # 红色实心圆
cv2.putText(img, "cat: 0.98", (60, 40),
            cv2.FONT_HERSHEY_SIMPLEX, 0.8, (0, 255, 0), 2)
# putText 只支持 ASCII——中文标注得先画到 PIL/Pillow 再转回 ndarray

cap = cv2.VideoCapture(0)              # 或 VideoCapture("demo.mp4")
while cap.isOpened():
    ok, frame = cap.read()             # ok=False 表示读完了
    if not ok:
        break
    cv2.imshow("gray", cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY))
    if cv2.waitKey(1) & 0xFF == ord("q"):   # 按 q 退出
        break
cap.release()
cv2.destroyAllWindows()
# 写视频用 cv2.VideoWriter(路径, fourcc, 帧率, (宽, 高))，注意宽高顺序

fourcc = cv2.VideoWriter_fourcc(*"mp4v")      # 编码器要与扩展名匹配
writer = cv2.VideoWriter("out.mp4", fourcc, 30.0, (1280, 720))

cap = cv2.VideoCapture(0)
cap.set(cv2.CAP_PROP_FRAME_WIDTH, 1280)      # 尺寸要和 Writer 声明一致
cap.set(cv2.CAP_PROP_FRAME_HEIGHT, 720)
while cap.isOpened():
    ok, frame = cap.read()
    if not ok:
        break
    writer.write(frame)
cap.release()
writer.release()                              # 不 release 文件会损坏

detector = cv2.CascadeClassifier(
    cv2.data.haarcascades + "haarcascade_frontalface_default.xml"
)
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
faces = detector.detectMultiScale(gray, scaleFactor=1.1, minNeighbors=5)
for (x, y, w2, h2) in faces:                  # 返回 [(x, y, w, h), ...]
    cv2.rectangle(img, (x, y), (x + w2, y + h2), (255, 0, 0), 2)

# 用法: python watermark.py ./photos ./out
# 依赖: pip install opencv-python numpy
import sys
from pathlib import Path

import cv2
import numpy as np

src, out = Path(sys.argv[1]), Path(sys.argv[2])
out.mkdir(parents=True, exist_ok=True)

for path in sorted(src.glob("*.jpg")):
    img = cv2.imread(str(path))
    if img is None:                       # 坏文件/乱路径：imread 静默返回 None
        print(f"跳过（读不了）: {path.name}")
        continue

    # 1. 右下角水印：先画黑色描边再画白字，任何背景都看得清
    h, w = img.shape[:2]
    cv2.putText(img, "AI Code Study", (w - 262, h - 26),
                cv2.FONT_HERSHEY_SIMPLEX, 0.9, (0, 0, 0), 3)
    cv2.putText(img, "AI Code Study", (w - 262, h - 26),
                cv2.FONT_HERSHEY_SIMPLEX, 0.9, (255, 255, 255), 1)

    # 2. 灰度 + Canny，统一转回三通道再横向拼接
    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    edges = cv2.Canny(gray, 100, 200)
    trio = np.hstack([
        img,
        cv2.cvtColor(gray, cv2.COLOR_GRAY2BGR),   # 通道数必须一致才能拼
        cv2.cvtColor(edges, cv2.COLOR_GRAY2BGR),
    ])
    trio = cv2.resize(trio, None, fx=0.4, fy=0.4) # 三张图并排，缩到合理大小

    cv2.imwrite(str(out / f"{path.stem}_wm.jpg"), img)
    cv2.imwrite(str(out / f"{path.stem}_trio.jpg"), trio)
    print(f"完成: {path.name} → 水印图 + 三联对比图")

$ mkdir -p photos && cp ~/Desktop/*.jpg photos/
$ python watermark.py photos out
完成: cat.jpg → 水印图 + 三联对比图
完成: street.jpg → 水印图 + 三联对比图
$ ls out/
cat_trio.jpg    cat_wm.jpg    street_trio.jpg    street_wm.jpg

layer = img.copy()
cv2.putText(layer, "AI Code Study", (w - 262, h - 26),
            cv2.FONT_HERSHEY_SIMPLEX, 0.9, (255, 255, 255), 3)
img = cv2.addWeighted(layer, 0.35, img, 0.65, 0)   # 字 35%，原图 65%

img = cv2.imread("photo.jpg")
if img is None:
    raise FileNotFoundError(f"读不到: {path}（检查相对路径与工作目录）")
print(img.shape)      # 到这里才安全

# OpenCV 4.x：返回 (contours, hierarchy)
contours, _ = cv2.findContours(bin_img, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)

# 注意：findContours 会修改输入图，喂给它的二值图要先 copy 或现场生成
# 入参必须是二值图：先 threshold / Canny，再 findContours

import matplotlib.pyplot as plt

plt.imshow(cv2.cvtColor(img, cv2.COLOR_BGR2RGB))   # 先转 RGB 再显示
plt.axis("off")
plt.show()