0

I am writing a code in which I require subtraction of the background part of a single image. All the methods demonstrated online are for video files. Is there any way of subtracting the background when we have only a single image available? For example consider this image:

This figure is prepared from the Stanford house number dataset

My question is, is there any way of segregating the numbers from there background so that we can detect them quite easily? The Stanford house number dataset is available here

Ricky
  • 635
  • 2
  • 5
  • 20

1 Answers1

1

You need numbers segmentation, not background. After you segmented numbers, just invert mask. Many neural networks can be trained for this task.

Andrey Smorodov
  • 10,649
  • 2
  • 35
  • 42
  • Thank you for your answer. Can you post some links describing number segmentation techniques? – Ricky Apr 28 '20 at 04:57
  • I know no trained networks for it, but you can train your own. You can use fcn-8 architecture for instance: https://github.com/pierluigiferrari/fcn8s_tensorflow or YOLO based https://github.com/ArtyZe/yolo_segmentation there a lot of architectures you can use. Google instance segmentation or semantic segmentation, But you'll of cause need to annotate your dataset. – Andrey Smorodov Apr 28 '20 at 06:57