Click the "Visual Learning for Beginners" above, select "Star" or "Top"
Heavyweight content delivered first time
Basic Knowledge1. Digital Image:A digital image, also known as a digital photo or pixel image, is a two-dimensional image represented by a finite number of digital pixel values. A digital image is obtained by digitizing an analog image and is composed of pixels as basic elements, which can be stored and processed using digital computers or digital circuits.2. Digital Image Processing Includes:Image digitization; image transformation; image enhancement; image restoration; image compression coding; image segmentation; image analysis and description; image recognition and classification.3. Components of a Digital Image Processing System:Input (acquisition); storage; output (display); communication; image processing and analysis.4. Steps from “Analog Image” to “Digital Image”:Acquisition of image information; storage of image information; processing of image information; transmission of image information; output and display of image information.5. What Does a Digital Image 1600×1200 Mean? What Does the Grayscale Range 0~255 Indicate?A digital image of 1600×1200 indicates a spatial resolution of 1600×1200 pixels; the grayscale range of 0~255 indicates the 256 levels of gray, representing the light and dark relationships of the image through different shades of gray, which is an 8-bit grayscale resolution.6. What Are the Two Processes of Image Digitization? How Do They Affect the Quality of Digitized Images?Sampling; quantization. Sampling transforms a spatially continuous image into discrete points; the higher the sampling frequency, the more realistic the restored image. Quantization converts the sampled pixel points into discrete numeric values; the number of different grayscale values in a digital image is called the grayscale level, and the higher the level, the clearer the image.7. What Factors Are Related to the Data Volume of a Digitized Image?Image resolution; sampling rate; sample values.8. What Is a Grayscale Histogram? What Are Its Applications? What Information Can You Obtain from a Grayscale Histogram?A grayscale histogram reflects the frequency relationship of each grayscale level pixel in an image. It can be used to: judge whether image quantization is appropriate; determine the threshold for image binarization; calculate the area of objects in the image; calculate the amount of information in the image.From the grayscale histogram, you can obtain:The histogram of a dark image is composed of several components on the left side with lower grayscale values; the histogram of a bright image tends to the right side with higher grayscale values; a histogram of a low-contrast image is narrow and concentrated in the middle of the grayscale levels; a histogram of a high-contrast image is widely distributed and evenly spread.9. What Is Point Processing? Which Algorithms You Learned Belong to Point Processing?In local processing, output values that depend only on pixel grayscale are called point processing. For example: image contrast enhancement, image binarization.10. What Is Local Processing? Which Algorithms You Learned Belong to Local Processing?In processing an input image, calculating an output pixel value based on pixel values in a small neighborhood of the input image is called local processing. For example: moving average smoothing method, spatial domain sharpening method.Image Enhancement/Spatial Domain11. What Is the Purpose of Image Enhancement?The purpose of image enhancement is to improve the visual effect of an image, purposefully enhancing the overall or local characteristics of an image for a given application scenario, making originally unclear images clearer or enhancing certain features of interest, expanding the differences between different objects in the image, suppressing uninteresting features, thereby improving image quality and enriching information, achieving strong image interpretation and recognition effects, and meeting the needs of certain feature analysis.12. What Is the Histogram of a Grayscale Image? Briefly Describe How It Can Be Used to Judge Image Quality?A grayscale histogram is defined as the statistical relationship between each grayscale level in a digital image and its frequency of occurrence. It can describe the overall characteristics of the image, such as the grayscale range of the image, the frequency of each grayscale level, the distribution of grayscale levels, the average brightness and contrast of the entire image, etc.13. What Are Common Image Enhancement Methods?Linear transformation of images; nonlinear changes of images; histogram equalization and specification of images.14. What Kind of Processing Does the “Average Template” Perform on Images? Write Out the 3×3 and 5×5 “Average Templates”?Suppress noise and improve image quality.
15. What Kind of Processing Does “Median Filtering” Perform on Images? How Is It Calculated?Median filtering is a method that sorts the grayscale values of pixels within a sliding window and replaces the grayscale value of the center pixel of the window with the median value. It is a nonlinear smoothing method that effectively suppresses impulse interference and salt-and-pepper noise while protecting edges from blurring.16. What Are Point Processing, Local Processing, and Global Processing? Which Processing Do 3×3 Average Template, Histogram Correction, and Grayscale Inversion Belong To?Point processing: Output values depend only on pixel grayscale (Histogram correction).Local processing: The output pixel value is determined by pixel values in a small neighborhood of the input image (Grayscale inversion).Global processing: The change in grayscale of a pixel in the image is related to all pixel grayscale values in the image (3×3 average template).Image Transformation/Frequency Domain Processing17. Common Image Transformation Algorithms:Geometric transformations of images (image distortion correction, image scaling: bilinear interpolation, rotation, stitching).Image transformations (Fourier, cosine, Walsh-Hadamard, K-L transformation, wavelet transformation).Image frequency domain processing (enhancement algorithms: high-frequency enhancement, homomorphic filtering; smoothing and denoising: low-pass filtering).18. Why Perform Image Transformation? What Image Processing Applications Do Various Transformations Have?Image transformation plays a crucial role in digital image processing and analysis, serving as a common and effective analytical tool. The purpose of image transformation is to facilitate image processing problems; it aids in feature extraction from images; and it conceptually enhances understanding of image information.19. Briefly Describe the Differences Between Spatial Domain Filtering and Frequency Domain Filtering.Spatial domain method: Directly operates on the pixel grayscale of the image. Common algorithms: Grayscale transformation of images; histogram correction (equalization, specification); smoothing and sharpening processing; color enhancement.Frequency domain method: Operates on the transformed values of the image in the transformation domain, then obtains the desired enhancement results through inverse transformation. Common algorithms: Low-pass filtering; high-frequency enhancement filtering; homomorphic filtering.20. Which Filters Are Used for Image Enhancement, Denoising, and Edge Detection in the Frequency Domain?Enhancement — Homomorphic filterDenoising — Low-pass filterEdge detection — High-pass filter21. Steps for Frequency Domain Processing of Images:a. Remove noise and improve the visual effect of the imageb. Highlight edges to facilitate recognition and processingImage RestorationCommon Image Transformation Algorithms:(1) Inverse filtering; (2) Wiener filtering; (3) Blind convolution22. What Is Image Restoration? What Is the Difference Between Image Restoration and Image Enhancement?Image degradation refers to the decline in image quality caused by imperfections in imaging systems, transmission media, and devices during formation, transmission, and recording of the image.Image restoration and image enhancement differ in that both aim to improve image quality, but image enhancement does not consider how the image has degraded. It merely enhances the visual effect through various techniques. Thus, image enhancement can disregard whether the enhanced image is distorted, as long as it looks good. On the other hand, image restoration requires knowledge of the mechanisms and processes of image degradation to find a corresponding inverse process calculation method, thereby obtaining the restored image. If an image has degraded, restoration must be performed before enhancement.23. Mention Several Types of Image Degradation:Image blur, distortion, noise, etc.24. What Is a Wiener Filter?It is a linear filter optimized by the least squares criterion, under certain constraints, its output minimizes the square of the difference from a given function. Through mathematical operations, it can ultimately be transformed into a problem of solving a drag-and-drop equation, utilizing the correlation characteristics of stationary random processes and spectral characteristics to filter signals mixed with noise.25. Mention Several Common Image Restoration Methods:Algebraic restoration methods: Unconstrained restoration; constrained least squares method.Frequency domain restoration methods: Inverse filtering restoration method; removal of blur caused by uniform motion; Wiener filtering restoration method.Image Compression CodingCommon Image Transformation Algorithms:(1) Huffman coding; (2) Arithmetic coding; (3) Predictive coding; (4) Transform coding.26. Why Can Images Be Compressed? (What Redundancies Exist in Digital Images?)Images can be compressed because redundancies exist in the data. In image compression, there are three basic data redundancies: coding redundancy; inter-pixel redundancy; visual redundancy.27. What Are Lossy and Lossless Compression?Lossless compression: It is the compression of the file itself, optimizing the data storage method of the file using some algorithm to represent repeated data information. The file can be completely restored without affecting the file content, and for digital images, it will not cause any loss of image details.Lossy compression: It involves altering the image itself, retaining more brightness information while merging hue and chroma information with surrounding pixels. The merging ratio varies, and the compression ratio also differs. Due to the reduction of information, the compression ratio can be very high, and the image quality will correspondingly decline.28. What Is the Basic Idea of the Huffman Coding Algorithm?It is to encode based on the probability of occurrence of source data symbols. The symbols that appear more frequently in the source data are assigned shorter codewords, while those that appear less frequently are assigned longer codewords, thereby achieving representation of the source data with the least possible code.29. What Are the Differences Between Lossless and Lossy Predictive Coding Algorithms? At Which Stage Do They Achieve Compression?In lossless (also known as lossless, error-free, information-preserving) coding, only the redundant data in the image is removed, and the reconstructed image after decoding is identical to the original image without any distortion.Lossy (also known as error-prone, distorted) coding refers to a situation where the reconstructed image differs from the original image, and cannot be precisely restored, but visually appears similar. It is a coding method that achieves high compression ratios.30. Briefly Describe the Basic Principles of Statistical Coding, Arithmetic Coding, Predictive Coding, and Transform Coding Algorithms.Statistical coding: Variable-length codes based on the probability distribution of the source, making the average code length very close to entropy.Arithmetic coding: Uses the joint probability of coding symbols to replace a string of input symbols with a single floating-point number.Predictive coding: Instead of directly coding the signal, it encodes the prediction error of the image. Essentially, it encodes new information to eliminate correlation and redundancy between adjacent pixels.Transform coding algorithm: It transforms the image from the spatial domain to the transformation domain coefficients, where energy is concentrated, and then encodes the transformation coefficients to achieve data compression.Image Edge Detection and SegmentationCommon Algorithms in Image Processing:(1) Edge detection: Canny operator, Laplacian operator, Sobel operator.(2) Hough transform for detecting lines and circles.(3) Image segmentation: Threshold segmentation (also known as binarization), region segmentation algorithms: region growing method, split-and-merge method, spatial clustering method.31. Mention Several Common Edge Detection Operators:Canny operator, Laplacian operator, Sobel operator.32. For the Following Four Templates, Which Processing Implements Smoothing Denoising (Low-pass Filter), Image Sharpening (Gradient Method), and Edge Detection (High-pass Filter)?
Additional:1. Grayscale Image vs. Color Image:Grayscale Image: A two-dimensional grayscale (brightness) function f(x,y).Color Image: Composed of three two-dimensional grayscale functions f(x,y). The three are RGB or HSV.2. Image Processing:From image to image: from an image to another image after processing.A decision: transforming an image into a non-image representation.3. Adjacent Pixels:Four Neighborhood:
Diagonal Neighborhood:
Eight Neighborhood:
4. Pixel Connection, Adjacency, and Connectivity:Adjacency: only considers the spatial relationship between pixels.Based on the metric relationship between pixels:Four neighborhoods can also be defined as:
Eight neighborhoods can also be defined as:
Connectivity: spatial relationship that satisfies adjacency.Grayscale relationship: whether the grayscale values meet a specific similarity criterion, represented by V.
4-Connection: Two pixels p and r take values in V and r is in the 4-neighborhood of p.8-Connection: Two pixels p and r take values in V and r is in the 8-neighborhood of p.m-Connection (Mixed Connection): Two pixels p and r take values in V, and one of the following conditions is satisfied:1. r is in the 4-neighborhood of p. 2. r is in the diagonal neighborhood of p and the 4-neighborhood of p and the 4-neighborhood of r have an empty intersection. As shown in the figure: e.g., in the color image, the left side is m-connection, and the right side is not. 
5. Distance Between Pixels:Three pixels p, q, r, with coordinates (x, y), (s, t), (u, v), the metric function is denoted as D.Properties of pixel distances:1>D(p,q)>=0 The distance between two pixels is always positive.2>D(p,q)=D(q,p) The distance does not depend on the choice of start and end points.3>D(p,R)<=D(p,q)+D(q,r) The shortest distance is along a straight line.Distance metric functions:1>Euclidean distance: The distance p is less than or equal to a certain value: a circle with p as the origin.
2>City block distance: The distance is less than or equal to a certain value: a diamond centered on p.
3>Chessboard distance: The distance p is less than or equal to a certain value: a square centered on p.
Download 1: OpenCV-Contrib Extension Module Chinese Tutorial
Reply "Extension Module Chinese Tutorial" in the "Visual Learning for Beginners" WeChat public account to download the first Chinese version of the OpenCV extension module tutorial available online, covering installation of extension modules, SFM algorithms, stereo vision, target tracking, biological vision, super-resolution processing, and more than twenty chapters of content.
Download 2: Python Visual Practical Projects 52 Lectures
Reply "Python Visual Practical Projects" in the "Visual Learning for Beginners" WeChat public account to download 31 visual practical projects including image segmentation, mask detection, lane line detection, vehicle counting, adding eyeliner, license plate recognition, character recognition, emotion detection, text content extraction, and face recognition, to help quickly learn computer vision.
Download 3: OpenCV Practical Projects 20 Lectures
Reply "OpenCV Practical Projects 20 Lectures" in the "Visual Learning for Beginners" WeChat public account to download 20 practical projects based on OpenCV to advance OpenCV learning.
Group Chat
Welcome to join the reader group of the public account to communicate with peers. Currently, there are WeChat groups for SLAM, 3D vision, sensors, autonomous driving, computational photography, detection, segmentation, recognition, medical imaging, GAN, algorithm competitions, etc. (will be gradually subdivided in the future). Please scan the WeChat ID below to join the group, and note: "nickname + school/company + research direction", for example: "Zhang San + Shanghai Jiao Tong University + Visual SLAM". Please follow the format; otherwise, you will not be approved. After successful addition, you will be invited to the relevant WeChat group based on your research direction. Please do not send advertisements in the group; otherwise, you will be removed. Thank you for your understanding~