Foreword
Last weekend, the entrance examination for 2020 ended. For myself, although there are many unsatisfactory places, I always feel that there is something to be left in each stage. If only two words are finally obtained, whether it is completed or not, it will be for me Bigger blow. I like to think of each study as a brick that builds my own future knowledge system. I continue to cast bricks and continue to cast bricks. I hope that I can be strong enough one day, and then use these bricks to build a building of ideas, not It was not until I needed to build a building that I learned how to cast bricks. I am sharing some notes and feelings here, which is a sorting out of what I learned in the previous stage. Of course, I also hope my words can help friends in need.
examination
This is an exam after all, and everyone is most concerned about the exam itself.
830 Digital Image Processing Reference Book: Gonzalez. Digital Image Processing. Third Edition
830 digital image processing total score: 150
830 Digital Image Processing Exam Time: 14: 00-17: 00 in the afternoon of the second day of postgraduate entrance examination
830 digital image processing question types: noun explanation (not tested in the past 10 years), short answer questions (in fact, an upgraded version of noun explanation), analysis questions, calculation questions, proof questions, design questions In the past few years, there were 9 short answer questions, with a total of 45 points, and this year, 10 questions with a total of 50 points. Then analyze, calculate, prove and design a total of 9 questions, this year a total of 100 points.
830 Digital Image Processing Exam Outline: The outline is almost the same every year
1、基本概念
1）数字图像处理的概念
2）数字图像处理的基本步骤
2、数字图像处理基础
1）图像感知和获取
2）图像取样与量化
3）像素间的一些基本关系
4）数字图像处理中所用的基本操作
3、灰度变换与空间滤波
1）灰度变换与空间滤波基础
2）基本的灰度变换函数
3）直方图处理
4）空间滤波基础
5）平滑空间滤波器
6）锐化空间滤波器
7）混合空间增强法
4、频率域滤波
1）傅立叶变换的基本知识
2）使用频率域滤波器平滑图像
3）使用频率域滤波器增强图像
4）实现
5、图像复原
1）图像退化/复原过程的模型
2）噪声模型
3）只存在噪声的复原——空间滤波
6、彩色图像处理
1）彩色基础
2）彩色模型
3）伪彩色图像处理
4）全彩色图像处理基础
5）彩色变换
6）平滑和锐化
7、形态学图像处理
1）膨胀与腐蚀
2）开闭运算
3）基本形态学算法，包括：边界提取、孔洞填充、连通分量的提取
8、图像分割
1）点、线和边缘检测
2）阈值处理
3）基于区域的分割
4）基于形态学分水岭的分割
9、表示与描述
1）表示，包括：链码、骨架
2）边界描绘子，包括：傅里叶描绘子、统计矩
3）区域描绘子，包括：拓扑描绘子、纹理
4）使用主分量进行描绘830 Digital Image Processing Review Blueprint: From the exam I experienced this year, the real questions of the past years are still the best guide. If you count the repetition of ideas, the repetition rate this year is more than 90%. And this year, I examined the very unpopular questions on the real questions more than 10 years ago. It shows that there is no absolute cold and no absolute heat. If you want to achieve good results, it is very necessary to review it completely. I have n’t seen it for many years. I ’m sure I wo n’t take the exam. Then I glanced at it. As a result, I took two questions that I thought were 25 points.
Question types and knowledge points in real questions
Each chapter in the outline can be tested with short answer questions
Chapter One (★★)
Simple concept, short answer
Chapter 2 (★★★)
Human eye brightness adaptation, sampling quantification, resolution, etc. have appeared in the form of short answer
Affine transformation + interpolation , translation, rotation, scaling, etc., often appear in real questions 10 years ago, but basically did not pass the test, and then took the test the previous year to analyze the calculation problem
Many operations in this chapter are the basis for later learning
Chapter Three (★★★★★)
Gray layering, short answer
Grayscale transformation:
Compare the effects of two grayscale transformation functions, general short answer questions
Contrast stretch of the piecewise function , general calculation analysis questions, this year's exam ( similar to the fourth question in 2013 )
Histogram processing:
The concept of histogram, the relationship between histogram and original graph, and the concept of histogram equalization and then the relationship between histogram equalization and histogram matching, the effect of multiple histogram equalization ( tested this year, the same as the short answer question 7 of 2013 ) General short answer questions
Histogram equalization operation : analysis of calculation questions, almost never absent on real questions over the years, and basically all are 15 points, of course , this year there is no absence (similar to 2014 question 6)
Derivation of Histogram Equilibrium Transfer Function and Derivation of Histogram Prescribed Transfer Function . I loved the exam more than 10 years ago, I have n’t tested it for many years, and I took it this year ( same as 2012 Question 7 )
Spatial filtering:
The operation of the template operation is actually the addition, subtraction, multiplication and division within 10, and the analysis calculation problem
Convolution and related operations, appeared in previous years, analysis and calculation
Unsharp shielding , analysis + short answer ( this year's short answer is tested, the basic steps are the same as 2016 short answer question 6 )
Chapter 4 (★★★★)
Principles of Fourier transform, some applied principles of Fourier transform, frequency domain filtering steps, some high-pass low-pass filters, etc., general short answer questions
Ringing effect, homomorphic filtering ( this year's short answer, basic steps, the same as the 2016 short answer question 9 ), high frequency emphasis , this year a spatial domain first-order sharpening operator is derived in the frequency domain , Analysis and calculation proof questions.
Chapter Five (★★★★★)
Reasons for image degradation , degradation process, restoration process, short answer separately, combined analysis
Spatial filtering:
Common noises, and the filters they use, are generally short-answered ( this year I tested the effect of median filtering on salt and pepper and Gaussian noise, short-answered, similar to the 5th question of the 2013 short-answered questions )
Commonly used filters, their principles, advantages and disadvantages, general short answer
The operation process of various filters , that is, template calculation, analysis and calculation questions, the real questions in the past years are almost never absent, and this year is absent in the analysis and calculation questions. ( But this year I accidentally took this point in the 10th question of the short answer question, the same as the 8th question in 2014, just using the 5X5 and 9X9 filters )
Chapter Six (★★★)
Color image, grayscale image, binary image concept difference, short answer questions
Color image model and application, short answer
False color, false color, true color concept, short answer
Color image processing, generally converted to gray-scale image processing, can appear as an element in the design problem.
Chapter Seven (★★★★)
Knowledge points in the outline, short answer + analysis and calculation
Expansion and corrosion
Open and close operation
Boundary extraction, hole filling, connected component extraction
Chapter Eight (★★★★★)
Point and edge detection:
Laplace operator derivation and frequency domain expression, short answer + analysis calculation
Gradient operator + common filter , short answer + analysis calculation ( this year, I forgot to analyze or short answer, the advantages and disadvantages of Robert and prewitt )
Log derivation and use, analysis
Canny operator, I haven't tested it before. I took a short answer this year and answered its steps
Edge connection:
The test is more about the Hough transform , short answer + analysis
Threshold processing:
Observe the histogram to get the threshold, analyze
Threshold for minimum error estimation, analysis and calculation (not tested for many years)
Global automatic threshold estimation and analysis
Otsu (otsu), I rarely saw this knowledge point survey in previous years, this year I took a short answer, test steps
Edge improvement threshold, analysis + short answer ( a short answer question was tested this year, the same as question 8 in 2015 )
Moving average, analysis , ( tested this point this year, the same as the 8th question in 2016 )
Variable threshold of local characteristics, analysis
Region segmentation:
Region growing, short answer + points analysis
Regional division and aggregation, ( tested this point this year, inspection steps ), analysis
Watershed, short answer + analysis
Chapter Nine (★★★★)
Most of the real questions over the years are short answer questions
Texture + symbiosis matrix ( tested this year, similar to the 2016 question 6, except that the position operator is different from the image size ), chain code, principal component analysis, there have been analysis calculation questions
Design question (15 points)
License plate location segmentation, phone number location segmentation, adding fog, lights or something
Moving target detection
Decolorize the color image, add part of the image that is too dark, and part of the image that is too bright, etc. ( This year's test is similar to the 2016 question 10 )
According to the knowledge of Chapter VIII or Chapter VII, design problems have appeared in previous areas such as regional growth and log.
Judging the type of object according to the size and color of the object
The design question and the analysis question have a great relationship. The author thinks that the design question is just a synthesis of multiple analysis question test points, and another unfamiliar scene.
Conclusion
The above is just a summary of some personal experiences, or the arrangement of what I think is important. The reason why experience is called experience is that it is just for everyone to see. Specifically, it depends on how you feel. If my analysis above is one-sided, incomplete, or wrong, I hope to point out that we will make progress together. The last thing I want to say is that there is no absolute cold nor absolute heat. To achieve good results, a comprehensive review is necessary.
