[actual combat] realize page distortion correction with OpenCV

Click on the above “ Xiaobai studies vision ”, Optional plus " Star standard " or “ Roof placement ”

 Heavy dry goods , First time delivery 

Preface

Today's content is about page distortion correction , In order to let everyone have a related concept , Let's preview the renderings first

---

Principle of page distortion correction

The purpose of the author writing this article is also very interesting , because The author's wife is a teacher , Her students often send her some picture homework , However, it is inevitable that due to various reasons such as photography technology and conditions , It makes it more difficult to correct homework , So the author for his wife , We worked out this plan .（ The magic of love ）

For some classic ideas of text correction （ for example Leptonica dewarping method and the CTM method） Break down the problem into ：

1. Break the text into lines

2.   Find the mapping or coordinate transformation that makes the lines parallel and horizontal

The author of this article also adopts a similar idea , Build a parameter model for the page （ Text ） Correction of , The main parameters considered in the model ：

• The rotation vector of the page in three-dimensional space  r Vector of peaceful shift t

• Specify the two curvatures of the page surface  α and β

• On the page n Vertical offset of a horizontal span

• For each span , In horizontal span m The horizontal offset of a point

For some of the above parameters , My understanding is that page distortion exists the transformation of rotation and offset , Because the text content is divided into several lines , So think of each line of text as a curve , Let's assume that it turns into a cubic spline after being twisted , By two coefficients a and b control . Vertical offset refers to the distance between each row after segmentation , Horizontal offset refers to setting points on each line , For subsequent correction and transformation processing .

（ Personal understanding , There may be a little deviation from the original ）

The next operation is very awesome , The author simulates the page , Set different parameters to observe the rules

Finally, the author found that when the set parameters are fixed , Each key point on the page can find a certain corresponding point in the plane of the image .

The red ones are the key points detected on the text , The blue is the point through the re projection of the model . The blue dots on the left are collinear （ Suppose the text is straight at first ）, The image on the right will determine the posture of the page / shape , Make the blue dot almost on the red dot .

---

Implementation details

The above is the general implementation principle , Next is the main steps for the author to realize the function ：

1、 Get the boundary of the page . The whole picture is not used here , Instead, it cleverly uses the internal text content and the approximate distance between the text and the boundary to determine the boundary of the page .

2、 Detect the outline of the text . Through adaptive threshold ——> inflation + corrosion ——> Connected region analysis +PCA To approximate the text

3、 Make the text span .

4、 Sample span . Generate representative points on each span .

5、 Create initial parameter estimation . Realize projection into the picture plane .

6、 Optimize ！ Reduce the error of re projection

7、 Remap image and threshold .

---

summary

This project is used before , Today I specially share it with you , Baidu search out few related content , Because I think sharing is quite meaningful , Due to my limited understanding of the content , There may be some deviations , But it really took a lot of effort to interpret .

It's not difficult to use , But because of the author's hypothesis , For texts that are not in this case , And the correction effect of excessive deformation is not very ideal . It's not easy to create , I hope I can watch it , Forward first hand , Due to the lack of message function , therefore , Only through the data feedback in the background can we know whether the content we share is really helpful to everyone , Everyone's reading is watching and forwarding , I am very grateful .

The good news ！

Xiaobai learns visual knowledge about the planet

Open to the outside world

 download 1：OpenCV-Contrib Chinese version of extension module 

 stay 「 Xiaobai studies vision 」 Official account back office reply ： Extension module Chinese course , You can download the first copy of the whole network OpenCV Extension module tutorial Chinese version , Cover expansion module installation 、SFM Algorithm 、 Stereo vision 、 Target tracking 、 Biological vision 、 Super resolution processing and other more than 20 chapters .


 download 2：Python Visual combat project 52 speak 
 stay 「 Xiaobai studies vision 」 Official account back office reply ：Python Visual combat project , You can download, including image segmentation 、 Mask detection 、 Lane line detection 、 Vehicle count 、 Add Eyeliner 、 License plate recognition 、 Character recognition 、 Emotional tests 、 Text content extraction 、 Face recognition, etc 31 A visual combat project , Help fast school computer vision .


 download 3：OpenCV Actual project 20 speak 
 stay 「 Xiaobai studies vision 」 Official account back office reply ：OpenCV Actual project 20 speak , You can download the 20 Based on OpenCV Realization 20 A real project , Realization OpenCV Learn advanced .


 Communication group 

 Welcome to join the official account reader group to communicate with your colleagues , There are SLAM、 3 d visual 、 sensor 、 Autopilot 、 Computational photography 、 testing 、 Division 、 distinguish 、 Medical imaging 、GAN、 Wechat groups such as algorithm competition （ It will be subdivided gradually in the future ）, Please scan the following micro signal clustering , remarks ：” nickname + School / company + Research direction “, for example ：” Zhang San  +  Shanghai Jiaotong University  +  Vision SLAM“. Please note... According to the format , Otherwise, it will not pass . After successful addition, they will be invited to relevant wechat groups according to the research direction . Please do not send ads in the group , Or you'll be invited out , Thanks for your understanding ~