Classifier-Sample-Extractor

Installation

• You need to install opencv, python and numpy for your operating sytsem (Windows, Linux - Ubuntu, Mac OSX)

• Clone this repository

    $ git clone https://github.com/fabianbormann/Classifier-Sample-Extractor.git 
  

• Copy images containing objects that you wish to detect to the data/ folder

• Replace the cascade.xml with your own classifier (current cascade.xml is the frontal eye detector by Shameem Hameed bundled with the opencv installation)

Setup

• Update the properties in sample_extractor.py line 71 cascade.detectMultiScale(image, scaleFactor=1.2, minNeighbors=10,minSize=(20,20), maxSize=(45,45)) the sizes should fit to your training samples
• My images use the name convention data/image-xxx.png so that your positives and negatives will be named 'xxx_yyy_zzz.png' yyy and zzz describes the position (center) of this subimage in the image-xxx.png. If you don't want to use this feature simply edit line 37 and 39.

Run

    $ python sample_extractor.py

Your classifier will run at a random image from your data folder. Every detected object will be presented by a white border.

If you click with the left mouse button the border will be change it's color to green and the subimage will move to the positives folder. In contrast if you click the right mouse button border color switch to red and the subimge move to negatives folder. The middle mouse button load the next image from the data folder.