IMAGE FEATURE RECOGNITION IN Qt / C++ FOR OpenCV EXPERTS - 27/02/2021 14:52 EST

公共说明面板

• 
  竞赛主办者

  • 3 年 之前

  it's possible that not all of the given code modules are performing quite well ... I found them on the internet but I didn't test them myself up to now (some examples attached seemed to work, but they should not be more then a starting point or placeholder ... )

  • 3 年 之前
• 
  竞赛主办者

  • 3 年 之前

  Two questions ?
  
  1.) Did you rely on the preprocessed versions?
  2.) Did you rely (basically) on the given codes?
  
  A) The contours of the glasses are found quite well !
  B) The side lines of the credit card ist found also quite well !
  C) The symmetry line of the contours is found well in picture 2 + 3 ! In picture 1 it seems not so exact?!
  D) The eye pupils are found relatively good in picture 3 ! In picture 1 + 3 they they seem not very exact ...
  E) Especially the top and bottom line of credit card seem to be a biger problem! I'm wondering about the reason? Perhaps here would help the preprocessed file ?!
  
  Thank you! Bye for now
  Best regards
  David

  • 3 年 之前
  查看 8消息
  1. 
     竞赛主办者

     • 3 年 之前

     I have some experience with image processing and also with C++ - openCV ... but I'm not used to Qt ... the main difficulty for me is to make a reasonable user interface and to set up all necessary libraries in Qt ... if I can base on a code/project template this helps me a lot ... my speciality are mathematical algorithms NOT coding user interfaces... if your entry will not be perfect in the functionality perhaps I can optimize it with your further help ... however your actual state (if coded in Qt/C++) will already be useful for me ... perhaps you will arrive to optimze it still a little in the next few days?

     • 3 年 之前
  2. 
     竞赛主办者

     • 3 年 之前

     in fact, the symmetry line has to be calculated from the inner contours of the glasses only ... not from the eyes ... I stated this in the text file of the contest, but I recognized now, that there is a misprint in my text: "contour D)" instead of "contour C)" ...

     • 3 年 之前
• 
  竞赛主办者

  • 3 年 之前

  Hello Oussama
  
  Thank you for your contribution. In fact you are the first with a reasonable contribution ... up to now !
  It does quite well so far!
  
  Did you recognize that I have added the pictures in a black & white version and in a preprocessed version ???
  
  I think especially the preprocessed version (on which you can rely) does it make much easier to find the lines!

  • 3 年 之前
• 
  竞赛主办者

  • 3 年 之前

  Okay Oussama I'll do add some time. Please note I have added preprocessed files where the contour lines are outlined. You can base on these images (combining with the original images). This makes the task much easier! A minimal requirement is to integrate the given sources in Qt and to make them compilable together with OpenCV. Even if the sources will not deliver good results or only partially.

  • 3 年 之前
  1. 

     gzperra

     • 3 年 之前

     Thanks. I’ve already integrated the sources in GIVEN OPENCV MODULES.zip. Unfortunately, they didn’t give acceptable results with all images. So I’ve created my own functions and I’ve done almost 60% of the work: eyes detection + glasses contours detection.. and without the use of the preprocessed files.

     • 3 年 之前
• 

  gzperra

  • 3 年 之前

  #extended
  Could you add 2 or 3 more days ? I've just started working yesterday.

  • 3 年 之前
  1. 
     竞赛主办者

     • 3 年 之前

     Okay Oussama I'll do add some time. Please note I have added preprocessed files where the contour lines are outlined. You can base on these images (combining with the original images). This makes the task much easier! A minimal requirement is to integrate the given sources in Qt and to make them compilable together with OpenCV. Even if the sources will not deliver good results or only partially.

     • 3 年 之前
• 
  竞赛主办者

  • 3 年 之前

  To make it easier I have added preprocessed states of the 3 examples which can be used in parallel or instead of the original examples.
  I have also added the black&white versions of the 3 color examples. It's also okay to use these pictures instead.

  • 3 年 之前
• 

  dataexpert18

  • 3 年 之前

  C++ is mandotry, can we use python instead of c++???

  • 3 年 之前
  1. 
     竞赛主办者

     • 3 年 之前

     C++ and Qt ARE requirements !
     The given codes to disposal are all in C++ as well.
     These codes have to be PRIMARILY adapted and integrated in a whole Qt-project as described!
     If this project ist compilable and buildable (with the given codes adapted to Qt!) exactly as described - even with still poor recognition rate - the first step is done successfully and could be submitted to the contest!
     A following optimization of this project (to WIN contest in first place) should also be in C++ ...
     If this in done Python (of course eventually quicker for "trial an error"-method) it has to be back-translated to C++ .... This should not be a serious problem because the corresponding OpenCV-calls exist in both languages!

     • 3 年 之前
• 
  竞赛主办者

  • 3 年 之前

  C++ and Qt ARE requirements !
  The given codes to disposal are all in C++ as well.
  These codes have to be PRIMARILY adapted and integrated in a whole Qt-project as described!
  If this project ist compilable and buildable (with the given codes adapted to Qt!) exactly as described - even with still poor recognition rate - the first step is done successfully and could be submitted to the contest!
  A following optimization of this project (to WIN contest in first place) should also be in C++ ...
  If this in done Python (of course eventually quicker for "trial an error"-method) it has to be back-translated to C++ .... This should not be a serious problem because the corresponding OpenCV-calls exist in both languages!

  • 3 年 之前

显示更多评论