【深入OpenCV图像处理：从基础到实战应用】

import cv2

# 高级图像读取参数详解
# 参数1：图像路径 | 参数2：读取模式(cv2.IMREAD_COLOR/cv2.IMREAD_GRAYSCALE) 
# 参数3：指定解码格式(如cv2.IMREAD_REDUCED_COLOR_2)
img = cv2.imread('input.jpg', cv2.IMREAD_COLOR)  

# 获取图像维度信息 (高度, 宽度, 通道数)
print(f"Image Shape: {img.shape}")  # 输出格式：(H, W, C)

# 带错误处理的图像显示方案
if img is not None:
    cv2.imshow('Demo', img)
    # waitKey参数为等待时间(ms)，0表示无限等待
    key = cv2.waitKey(0)  
    # 按ESC键退出(ASCII 27)
    if key == 27:  
        cv2.destroyAllWindows()
else:
    print("Error: Image loading failed!")

# 对比度受限自适应直方图均衡化(CLAHE)
clahe = cv2.createCLAHE(
    clipLimit=2.0,  # 对比度限制阈值
    tileGridSize=(8,8)  # 局部直方图区域划分
)
enhanced_img = clahe.apply(gray_img)

# 结构元素定义
kernel = cv2.getStructuringElement(
    shape=cv2.MORPH_ELLIPSE,  # 形状类型(MORPH_RECT/MORPH_CROSS)
    ksize=(5,5)  # 核尺寸
)

# 闭运算（先膨胀后腐蚀）
closed_img = cv2.morphologyEx(
    src=img, 
    op=cv2.MORPH_CLOSE, 
    kernel=kernel,
    iterations=3  # 操作次数
)

def defect_detection_pipeline(img_path):
    # 1. 图像预处理
    img = cv2.imread(img_path)
    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    
    # 2. 噪声抑制（中值滤波）
    denoised = cv2.medianBlur(gray, 5)
    
    # 3. 边缘增强（Sobel算子）
    sobel_x = cv2.Sobel(denoised, cv2.CV_64F, 1, 0, ksize=3)
    
    # 4. 阈值分割（大津法）
    _, thresh = cv2.threshold(
        src=sobel_x, 
        thresh=0, 
        maxval=255, 
        type=cv2.THRESH_BINARY + cv2.THRESH_OTSU
    )
    
    # 5. 形态学优化
    kernel = np.ones((3,3), np.uint8)
    opened = cv2.morphologyEx(thresh, cv2.MORPH_OPEN, kernel)
    
    # 6. 缺陷区域标记
    contours, _ = cv2.findContours(
        image=opened,
        mode=cv2.RETR_EXTERNAL,  # 仅检测外轮廓
        method=cv2.CHAIN_APPROX_SIMPLE
    )
    
    # 绘制检测结果
    result = img.copy()
    cv2.drawContours(result, contours, -1, (0,0,255), 2)
    return result

# SIFT特征检测器配置
sift = cv2.SIFT_create(
    nfeatures=0,        # 保留的特征点数量(0表示无限制)
    nOctaveLayers=3,    # 金字塔层数
    contrastThreshold=0.04,  # 对比度阈值
    edgeThreshold=10    # 边缘阈值
)

# 关键点检测与描述
keypoints, descriptors = sift.detectAndCompute(gray_img, None)

# 可视化特征点
vis_img = cv2.drawKeypoints(
    image=img, 
    keypoints=keypoints, 
    outImage=None,
    flags=cv2.DRAW_MATCHES_FLAGS_DRAW_RICH_KEYPOINTS
)

# 模型加载
net = cv2.dnn.readNetFromONNX("yolov5s.onnx")

# 输入预处理
blob = cv2.dnn.blobFromImage(
    image=img, 
    scalefactor=1/255.0,  # 归一化系数
    size=(640, 640),      # 输入尺寸
    mean=(0,0,0),         # 均值减法
    swapRB=True,          # BGR转RGB
    crop=False            # 中心裁剪
)

# 前向推理
net.setInput(blob)
outputs = net.forward(net.getUnconnectedOutLayersNames())

# 后处理（示例代码）
for detection in outputs[0][0]:
    confidence = detection[4]
    if confidence > 0.5:
        x, y, w, h = detection[0:4] * np.array([img_w, img_h, img_w, img_h])
        cv2.rectangle(img, (int(x-w/2), int(y-h/2)), (int(x+w/2), int(y+h/2)), (0,255,0), 2)

def ocr_preprocessing(image):
    # 1. 透视变换矫正
    gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
    blurred = cv2.GaussianBlur(gray, (5,5), 0)
    edged = cv2.Canny(blurred, 75, 200)
    
    # 2. 文档轮廓检测
    cnts = cv2.findContours(edged.copy(), cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE)
    cnts = imutils.grab_contours(cnts)
    cnts = sorted(cnts, key=cv2.contourArea, reverse=True)[:5]
    
    # 3. 透视变换矩阵计算
    screenCnt = None
    for c in cnts:
        peri = cv2.arcLength(c, True)
        approx = cv2.approxPolyDP(c, 0.02*peri, True)
        if len(approx) == 4:
            screenCnt = approx
            break
    
    # 4. 执行透视变换
    warped = four_point_transform(gray, screenCnt.reshape(4,2))
    
    # 5. 二值化处理
    thresh = cv2.threshold(warped, 0, 255, cv2.THRESH_BINARY | cv2.THRESH_OTSU)[1]
    return thresh