OpenCV Image Processing Complete Guide | OpenCV图像处理完全指南

Introduction: The Inner Logic of Image Processing | 引言：图像处理的内在逻辑

Fundamental Concepts | 核心概念

In computer vision, images are represented as matrices of pixel values. A color image typically consists of three channels (BGR in OpenCV), while a grayscale image has a single channel. Each pixel value ranges from 0 to 255, representing intensity.

在计算机视觉中，图像被表示为像素值矩阵。彩色图像通常包含三个通道（OpenCV中为BGR格式），而灰度图像只有一个通道。每个像素值的范围是0到255，表示亮度强度。

Processing Pipeline | 处理流程

The operations in this guide follow a logical sequence:

本指南中的操作遵循以下逻辑流程：

1. Image Acquisition → Reading images/videos from files or cameras 图像获取 → 从文件或摄像头读取图像/视频
2. Preprocessing → Basic operations like resizing, cropping, color space conversion 预处理 → 调整大小、裁剪、颜色空间转换等基本操作
3. Enhancement → Smoothing, sharpening, contrast adjustment 增强 → 平滑、锐化、对比度调整
4. Feature Extraction → Edge detection, contour finding, gradient calculation 特征提取 → 边缘检测、轮廓查找、梯度计算
5. Transformation → Pyramids, Fourier transform for frequency domain analysis 变换 → 金字塔、傅里叶变换用于频域分析

Core Principles | 核心原理

• Convolution (卷积): The foundation of many operations (smoothing, edge detection). A kernel slides over the image, computing weighted sums. 卷积：许多操作的基础（平滑、边缘检测）。一个核在图像上滑动，计算加权和。
• Thresholding (阈值化): Converts grayscale images to binary by comparing pixel values to a threshold. 阈值化：通过将像素值与阈值比较，将灰度图像转换为二值图像。
• Morphology (形态学): Based on set theory, operations like erosion and dilation modify shape. 形态学：基于集合论，腐蚀和膨胀等操作修改形状。
• Frequency Domain (频域): Transforms spatial information to frequency domain for filtering. 频域：将空间信息转换到频域进行滤波。

Why These Operations? | 为什么需要这些操作？

These operations form the building blocks of computer vision systems. They enable:

这些操作构成了计算机视觉系统的基础模块，使我们能够：

• Object Detection (目标检测): Identify and locate objects in images 目标检测：识别并定位图像中的物体
• Image Segmentation (图像分割): Separate foreground from background 图像分割：分离前景和背景
• Feature Matching (特征匹配): Compare and match patterns across images 特征匹配：在图像间比较和匹配模式
• Image Restoration (图像恢复): Remove noise and artifacts 图像恢复：去除噪声和伪影

Understanding these operations is essential for developing robust computer vision applications, especially for RM (RoboMaster) competition scenarios where real-time processing and accuracy are critical.

理解这些操作对于开发强大的计算机视觉应用至关重要，特别是在RM（RoboMaster）竞赛场景中，实时处理和准确性至关重要。

Version Compatibility | 版本兼容性

This guide is based on OpenCV 4.x. Note that:

• OpenCV 4.x: cv2.findContours() returns (contours, hierarchy)
• OpenCV 3.x: cv2.findContours() returns (image, contours, hierarchy)

本指南基于OpenCV 4.x编写。注意：

• OpenCV 4.x：cv2.findContours()返回(contours, hierarchy)
• OpenCV 3.x：cv2.findContours()返回(image, contours, hierarchy)

Table of Contents | 目录

1. Basic Image Operations
   • Reading Images
   • Displaying Images
   • Saving Images
   • Image Properties
   • Reading Video
   • ROI Extraction
   • Color Channel Operations
   • Border Padding
   • Arithmetic Operations
   • Image Blending
   • Image Resizing
2. Image Transformations
   • Color Space Conversion
   • Image Thresholding
   • Image Smoothing
   • Morphological Operations
   • Image Gradient
   • Edge Detection
   • Image Pyramid
   • Image Contours
   • Fourier Transform
3. Practical Tips & FAQ

1. Basic Image Operations | 图像基本操作

1.1 Reading Images (cv2.imread) | 图像读取

flags Parameter Details (flags参数详解):

Note (注意): OpenCV reads images in BGR format, not RGB! When displaying with Matplotlib, convert to RGB using cv2.cvtColor(img, cv2.COLOR_BGR2RGB).

OpenCV读取图像的格式是BGR，不是RGB！用Matplotlib显示时，需要用cv2.cvtColor(img, cv2.COLOR_BGR2RGB)转换。

import cv2

img = cv2.imread('cat.jpg')  # Read color image (BGR format) | 读取彩色图像（BGR格式）
img_gray = cv2.imread('cat.jpg', cv2.IMREAD_GRAYSCALE)  # Read grayscale | 读取灰度图

1.2 Displaying Images (cv2.imshow) | 图像显示

Supporting Functions (配合使用的函数):

cv2.imshow('image', img)  # Display image | 显示图像
cv2.waitKey(0)  # Press any key to continue | 按任意键继续
cv2.destroyAllWindows()  # Close windows | 关闭窗口

# Convenience function | 便捷函数
def cv_show(name, img):
    cv2.imshow(name, img)
    cv2.waitKey(0)
    cv2.destroyAllWindows()

1.3 Saving Images (cv2.imwrite) | 图像保存

cv2.imwrite('my_cat.png', img)  # Save as PNG | 保存为PNG格式

1.4 Image Properties | 图像属性

print(img.shape)   # (414, 500, 3)
print(img.size)    # 621000
print(img.dtype)   # uint8
print(type(img))   # <class 'numpy.ndarray'>

1.5 Reading Video (cv2.VideoCapture) | 视频读取

Common Methods (常用方法):

vc = cv2.VideoCapture('test.mp4')  # Open video file | 打开视频文件

if not vc.isOpened():
    print("Cannot open video")  # Cannot open video | 无法打开视频
    exit()

while True:
    ret, frame = vc.read()  # Read frame | 读取帧
    if not ret:
        break
    gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
    cv2.imshow('result', gray)
    if cv2.waitKey(100) & 0xFF == 27:  # ESC to exit | ESC键退出
        break

vc.release()
cv2.destroyAllWindows()

1.6 ROI Extraction | ROI截取

ROI (Region of Interest) | ROI（感兴趣区域）

img = cv2.imread('cat.jpg')
cat_face = img[0:50, 0:200]  # [row range, column range] | [行范围, 列范围]
cv_show('cat_face', cat_face)

1.7 Color Channel Operations | 颜色通道操作

Channel Split (cv2.split) | 通道分离

Note (注意): Order is B, G, R because OpenCV uses BGR format. 返回顺序是B、G、R，因为OpenCV使用BGR格式。

Channel Merge (cv2.merge) | 通道合并

b, g, r = cv2.split(img)  # Split channels | 分离通道
img = cv2.merge((b, g, r))  # Merge channels | 合并通道

# Keep only red channel | 只保留红色通道
cur_img = img.copy()
cur_img[:, :, 0] = 0  # Set B channel to 0 | B通道置0
cur_img[:, :, 1] = 0  # Set G channel to 0 | G通道置0

1.8 Border Padding (cv2.copyMakeBorder) | 边界填充

borderType Options (borderType填充类型):

top_size, bottom_size, left_size, right_size = (50, 50, 50, 50)

replicate = cv2.copyMakeBorder(img, top_size, bottom_size, left_size, right_size, cv2.BORDER_REPLICATE)
reflect = cv2.copyMakeBorder(img, top_size, bottom_size, left_size, right_size, cv2.BORDER_REFLECT)
reflect101 = cv2.copyMakeBorder(img, top_size, bottom_size, left_size, right_size, cv2.BORDER_REFLECT_101)
wrap = cv2.copyMakeBorder(img, top_size, bottom_size, left_size, right_size, cv2.BORDER_WRAP)
constant = cv2.copyMakeBorder(img, top_size, bottom_size, left_size, right_size, cv2.BORDER_CONSTANT, value=0)

1.9 Arithmetic Operations | 数值计算

Direct Arithmetic (numpy) | 直接算术运算（numpy方式）

img_cat = cv2.imread('cat.jpg')
img_cat2 = img_cat + 10  # Add 10 to each pixel | 每个像素加10

# Numpy wraps around on overflow | numpy加法溢出取模
result = img_cat + img_cat2  # 256 becomes 0, 257 becomes 1

OpenCV Addition (cv2.add) | OpenCV加法

Feature (特点): Clamps values at 255 (saturation operation) | 超过255时取255（饱和操作）

result = cv2.add(img_cat, img_cat2)  # Values > 255 become 255 | 超过255变为255

1.10 Image Blending (cv2.addWeighted) | 图像融合

Formula (公式): dst = src1 * alpha + src2 * beta + gamma

img_cat = cv2.imread('cat.jpg')
img_dog = cv2.imread('dog.jpg')

# Resize to same size | 调整到相同大小
img_dog = cv2.resize(img_dog, (500, 414))

# Blend: cat 40%, dog 60% | 融合：cat占40%，dog占60%
res = cv2.addWeighted(img_cat, 0.4, img_dog, 0.6, 0)

1.11 Image Resizing (cv2.resize) | 图像缩放

res = cv2.resize(img, (500, 400))  # Specify target size | 指定目标大小
res = cv2.resize(img, (0, 0), fx=2, fy=2)  # Double size | 放大2倍
res = cv2.resize(img, (0, 0), fx=0.5, fy=0.5)  # Half size | 缩小一半

2. Image Transformations | 图像变换处理

2.1 Color Space Conversion (cv2.cvtColor) | 颜色空间转换

Common Conversion Codes (常用转换代码):

2.1.1 Grayscale Conversion | 灰度转换

img_gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

2.1.2 HSV Color Space | HSV颜色空间

HSV Channel Meanings (HSV三通道含义):

Why HSV for RM? (为什么RM常用HSV?)

• RGB对光照敏感，同样颜色在亮暗环境下R/G/B值变化大
• HSV更鲁棒：H通道只表示颜色种类，光照变化影响小
• HSV对光照更鲁棒，H通道色调不受亮度影响

hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)

# Extract red region | 提取红色区域
lower_red = np.array([0, 100, 100])
upper_red = np.array([10, 255, 255])
mask = cv2.inRange(hsv, lower_red, upper_red)

2.2 Image Thresholding (cv2.threshold) | 图像阈值

Five Threshold Types (五种阈值类型):

Visual Diagram (可视化图示):

Original:     ████████████████████
BINARY:       ████████████████▓▓▓▓  (>127变白)
BINARY_INV:   ▓▓▓▓███████████████  (>127变黑)
TRUNC:        ████████████████████  (>127变127)
TOZERO:       ??????????██████████  (<127变黑)

ret, thresh1 = cv2.threshold(img_gray, 127, 255, cv2.THRESH_BINARY)
ret, thresh2 = cv2.threshold(img_gray, 127, 255, cv2.THRESH_BINARY_INV)
ret, thresh3 = cv2.threshold(img_gray, 127, 255, cv2.THRESH_TRUNC)
ret, thresh4 = cv2.threshold(img_gray, 127, 255, cv2.THRESH_TOZERO)
ret, thresh5 = cv2.threshold(img_gray, 127, 255, cv2.THRESH_TOZERO_INV)

2.3 Image Smoothing | 图像平滑

2.3.1 Mean Filter (cv2.blur) | 均值滤波

Feature (特点): Simple average, fast but blurs edges | 简单平均，速度快，但边缘模糊严重

blur = cv2.blur(img, (3, 3))

2.3.2 Box Filter (cv2.boxFilter) | 方框滤波

box = cv2.boxFilter(img, -1, (3, 3), normalize=True)  # Equivalent to blur | 等价于blur

2.3.3 Gaussian Filter (cv2.GaussianBlur) | 高斯滤波

Feature (特点): Gaussian weights, better edge preservation | 高斯权重，边缘保留较好

gaussian = cv2.GaussianBlur(img, (5, 5), 1)

2.3.4 Median Filter (cv2.medianBlur) | 中值滤波

Feature (特点): Uses median value, best for salt-and-pepper noise | 用中值替代，对椒盐噪声效果最好

median = cv2.medianBlur(img, 5)

Comparison (对比):

2.4 Morphological Operations | 形态学操作

2.4.1 Erosion (cv2.erode) | 腐蚀

Effect (效果): Shrinks foreground, removes small noise | 前景缩小，去除毛刺

kernel = np.ones((3, 3), np.uint8)
erosion = cv2.erode(img, kernel, iterations=1)

2.4.2 Dilation (cv2.dilate) | 膨胀

Effect (效果): Expands foreground, fills holes | 前景扩大，填补空洞

dilation = cv2.dilate(img, kernel, iterations=1)

2.4.3 Advanced Operations (cv2.morphologyEx) | 高级形态学操作

kernel = np.ones((5, 5), np.uint8)
opening = cv2.morphologyEx(img, cv2.MORPH_OPEN, kernel)  # 开运算
closing = cv2.morphologyEx(img, cv2.MORPH_CLOSE, kernel)  # 闭运算
gradient = cv2.morphologyEx(img, cv2.MORPH_GRADIENT, kernel)  # 梯度

2.5 Image Gradient | 图像梯度

2.5.1 Sobel Operator (cv2.Sobel) | Sobel算子

Important (重要): Sobel计算时白到黑是负数，会被截断为0。解决：取绝对值 + convertScaleAbs

正确做法：分离计算x和y方向，然后加权融合

# Separate calculation | 分离计算
sobelx = cv2.Sobel(img, cv2.CV_64F, 1, 0, ksize=3)
sobelx = cv2.convertScaleAbs(sobelx)  # Take absolute value | 取绝对值

sobely = cv2.Sobel(img, cv2.CV_64F, 0, 1, ksize=3)
sobely = cv2.convertScaleAbs(sobely)

# Combine | 融合
sobelxy = cv2.addWeighted(sobelx, 0.5, sobely, 0.5, 0)

2.5.2 Scharr Operator (cv2.Scharr) | Scharr算子

Feature (特点): More sensitive than Sobel, fixed ksize=3 | 比Sobel更灵敏，ksize固定为3

scharrx = cv2.Scharr(img, cv2.CV_64F, 1, 0)
scharry = cv2.Scharr(img, cv2.CV_64F, 0, 1)

2.5.3 Laplacian Operator (cv2.Laplacian) | Laplacian算子

Feature (特点): Second derivative, noise sensitive, usually needs prior smoothing | 二阶导数，对噪声敏感，通常需要先滤波

blurred = cv2.GaussianBlur(img, (3, 3), 0)  # Prior smoothing | 先去噪
laplacian = cv2.Laplacian(blurred, cv2.CV_64F)
laplacian = cv2.convertScaleAbs(laplacian)

2.6 Edge Detection (cv2.Canny) | 边缘检测

Canny Algorithm Steps (Canny算法步骤):

1. Gaussian blur for noise reduction (高斯滤波去噪)
2. Calculate gradient magnitude and direction (计算梯度)
3. Non-maximum suppression (非极大值抑制)
4. Double threshold detection (双阈值检测)
5. Edge tracking by hysteresis (抑制孤立弱边缘)

Threshold Selection Guide (阈值选择指南):

• threshold2 ≈ 2~3倍 threshold1 | 最佳平衡
• 阈值都低 → 检测多（可能含噪声）
• 阈值都高 → 检测少（可能丢失真正边缘）
• 调试建议：先从(50, 150)开始，根据结果调整

v1 = cv2.Canny(img, 50, 100)   # Loose (宽松)
v2 = cv2.Canny(img, 80, 150)   # Moderate (适中) - commonly used for RM
v3 = cv2.Canny(img, 100, 200)  # Strict (严格)

2.7 Image Pyramid | 图像金字塔

2.7.1 Up-sampling (cv2.pyrUp) | 向上采样

Effect (效果): Image size doubles (图像放大2倍)

up = cv2.pyrUp(img)

2.7.2 Down-sampling (cv2.pyrDown) | 向下采样

Effect (效果): Image size halves (图像缩小2倍)

down = cv2.pyrDown(img)

2.7.3 Laplacian Pyramid | 拉普拉斯金字塔

# Laplacian = Original - pyrUp(pyrDown(Original))
down = cv2.pyrDown(img)
down_up = cv2.pyrUp(down)
laplacian = img - down_up

2.8 Image Contours | 图像轮廓

2.8.1 Find Contours (cv2.findContours) | 查找轮廓

OpenCV Version Note (版本注意):

• OpenCV 4.x: returns (contours, hierarchy)
• OpenCV 3.x: returns (image, contours, hierarchy)

Retrieval Modes (检索模式):

Approximation Methods (逼近方法):

ret, thresh = cv2.threshold(img_gray, 127, 255, cv2.THRESH_BINARY)
contours, hierarchy = cv2.findContours(thresh, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)

2.8.2 Draw Contours (cv2.drawContours) | 绘制轮廓

draw_img = img.copy()
res = cv2.drawContours(draw_img, contours, -1, (0, 0, 255), 2)  # -1 = all contours

2.8.3 Contour Features | 轮廓特征

cnt = contours[0]
area = cv2.contourArea(cnt)  # Area | 面积
perimeter = cv2.arcLength(cnt, True)  # Perimeter | 周长

# Polygon approximation | 多边形逼近
epsilon = 0.15 * perimeter  # 0.01-0.1 for fine to rough | 精细到粗糙
approx = cv2.approxPolyDP(cnt, epsilon, True)

# Bounding rectangle | 边界矩形
x, y, w, h = cv2.boundingRect(cnt)

# Minimum enclosing circle | 外接圆
(x, y), radius = cv2.minEnclosingCircle(cnt)

2.9 Fourier Transform | 傅里叶变换

2.9.1 DFT (cv2.dft) | 离散傅里叶变换

Key Concepts (核心概念):

import numpy as np

img = cv2.imread('lena.jpg', 0)
img_float = np.float32(img)  # Must be float32 | 必须是浮点型

# DFT | 傅里叶变换
dft = cv2.dft(img_float, flags=cv2.DFT_COMPLEX_OUTPUT)
dft_shift = np.fft.fftshift(dft)  # Center low frequency | 将低频移到中心

# Magnitude spectrum | 幅值谱
magnitude = 20 * np.log(cv2.magnitude(dft_shift[:, :, 0], dft_shift[:, :, 1]))

2.9.2 IDFT (cv2.idft) | 逆傅里叶变换

Low-pass Filter Example (低通滤波示例):

rows, cols = img.shape
crow, ccol = rows // 2, cols // 2  # Center | 中心

# Create low-pass mask | 创建低通掩码
mask = np.zeros((rows, cols, 2), np.uint8)
mask[crow-30:crow+30, ccol-30:ccol+30] = 1  # Keep central region | 保留中心区域

# Apply mask | 应用掩码
fshift = dft_shift * mask

# Inverse shift | 逆中心化
f_ishift = np.fft.ifftshift(fshift)

# Inverse DFT | 逆变换
img_back = cv2.idft(f_ishift)
img_back = cv2.magnitude(img_back[:, :, 0], img_back[:, :, 1])  # Get magnitude | 取模

3. Practical Tips & FAQ | 实用技巧与常见问题

Performance Tips for RM | RM性能优化建议

Common Issues & Solutions | 常见问题与解决

Troubleshooting Guide | 故障排除指南

Q: Image reading returns None? | 图像读取返回None？ A: Check if the file path is correct and the file exists. 检查文件路径是否正确，文件是否存在。

Q: How to handle 4-channel images (RGBA)? | 如何处理4通道图像？ A: Use cv2.IMREAD_UNCHANGED to preserve alpha channel, or convert to 3-channel BGR. 使用cv2.IMREAD_UNCHANGED保留alpha通道，或转为3通道BGR。

Q: Morphology results are unexpected? | 形态学结果不符合预期？ A: Try different kernel sizes and iterations. Start with (3,3) and iterations=1. 尝试不同的核大小和迭代次数。从(3,3)和iterations=1开始。

Appendix: Quick Reference | 附录：常用函数速查

Basic Operations | 基础操作

Color Space | 颜色空间

Image Processing | 图像处理

Contours & Features | 轮廓与特征

Transformations | 变换

Document Version | 文档版本: 1.0 Based on | 基于: OpenCV 4.x Last Updated | 最后更新: 2026-05-03