-3

In the paper "Empirical analysis of detection cascades of boosted classifiers for rapid object detection", section 2.3, the authors talk about fast contrast stretching. I do not see in OpenCV or Matlab any reference to this preprocessing. Normally one has to insert variance normalized subwindows in the cascade classifier, but in the paper it is indicated that the image is normalized by an arbitrary constant which multiplies the standard deviation. I do not know if this constant (c=2 in the paper) was actually taken into account for training the cascade feature thresholds and if not, how could i use this operation to avoid another more complicated contrast enhancement technique in the detection process. Any hint would be very useful. Thanks

Question Edit: In the OpenCV Cascade Classification "Code", where exactly is applied the Contrast Stretching Constant (c=2) than Lienhart used when creating the cascades and that is described in the paper cited here http://docs.opencv.org/modules/objdetect/doc/cascade_classification.html? I can not find it. Thanks

satideur
  • 1
  • 2
  • 2
    Welcome to Stack Overflow. Questions here needs to be about specific programming issues with a clear definition of a problem and expected output, so someone knowledgeable can offer a valid solution. I am afraid your question is not a good format for this site as it is too broad and can draw too many opinionated answer. – Hoki Jul 31 '15 at 16:49
  • Thanks, I left only the programming issue – satideur Aug 01 '15 at 18:11

1 Answers1

0

You need to study the paper carefully. The paper is very clear and specific in terms of its details. In section 2.3, the contrast stretching is applied on a window. As a matter of fact all the computations described in the paper are with respect to a certain window which has to be analysed for the presence/absence of the object. It is used to do contrast stretching of the window before feeding it to the classifier.

PS : As @Hoki mentions, please avoid asking non-programming questions on StackOverflow.

Ujjwal Aryan
  • 3,827
  • 3
  • 20
  • 31