-1

I'm looking for the answers for choosing the background of the object detection . Now I'm working on the project of detecting the items using computer vision and , I'm encountering the problem in choosing the background for its , so what the background color should I choose for it in order to reach the best solution , I research about black , white , green ? Thank you !

Trần Hoàng Anh
  • 7
  • 1
  • Can you provide some information regarding the nature of these items? Especially their color would be interesting, as you don't want to camouflage them. – Alexander Mayer Jul 31 '19 at 06:45
  • for example , they are red , silver , or items grabbed by the nilon, they reflect with the light too much , or some items made by metal . Thank you ! – Trần Hoàng Anh Aug 01 '19 at 03:57
  • now i'm using the the black background ! , so is it good ? can you suggest for me other color background you think it's better ? – Trần Hoàng Anh Aug 01 '19 at 03:58
  • Please change your answers' title to resemble your exact problem - something along the lines of 'Which background color should I use to optimize my CV application?'. In addition, please add your object coloration information to the question. – Alexander Mayer Aug 01 '19 at 05:42

1 Answers1

0

Black or white should work as they - in addition to having a homogeneous coloration in the background - provide a way to capture noise or unwanted gradients when taking a look at the background only.

I name black and white specifically for their RGB/BGR values. If your objects are color X, you'll want to avoid color X as the background. Keep in mind that your image will (in the case of RGB/BGR) be three matrices with values ranging from 0 to the max value specified by your data type, i.e. 1 (if float) or 255 (if uint8). You want a background that is distinctively different from your object of interest.

Depending on your specific use case, additional information may or may not be relevant to you. If you are going for cascaded classifiers, the answer to this question contradicts my point. And this one shows some methods for detecting same-colored objects.

The answer to this one might also be interesting for you.

Alexander Mayer
  • 316
  • 2
  • 11