I have written a research paper in image processing that I will be submitting to an IEEE conference. I have proposed a new method to detect 'X' in images. However, due to hardware failure I could acquire only 4 sets of images. Thus, I have applied my method only on 4 images and I have also compared the results of previous two methods in literature by applying them to the same 4 images. The results obtained by my method are much better.

My question is, is it OK if I submit the paper showing results only on 4 images? The previous two papers have demonstrated their method on around 10 images or less. Also, is it OK to quantitatively compare the results (accuracy in percent) with just four images?

  • Did you make any use of those four images in developing your method? Commented Jan 31, 2016 at 18:08
  • @PatriciaShanahan: We had an idea and we validated our approach on those four images.
    – kunal18
    Commented Jan 31, 2016 at 18:11
  • 3
    "due to hardware failure I could acquire only 4 sets of images". I do not understand this. Get a friend of yours download additional sets and use a USB to get them on your PC.
    – Alexandros
    Commented Jan 31, 2016 at 18:21
  • @Alexandros: The 4 sets of experimental images were obtained using a special camera setup. The same camera setup cannot be used now because of some hardware failure.
    – kunal18
    Commented Jan 31, 2016 at 18:24
  • Email the authors of similar papers to send you their datasets then. If you are polite, they might accept.
    – Alexandros
    Commented Jan 31, 2016 at 18:25

1 Answer 1


The amount of experimental evidence that is needed in order to adequately demonstrate a result depends strongly on the current state of the art that you are comparing against.

In some cases, even a single instance will be interesting and publishable, as there is much knowledge that can be gleaned from it. See, for example, the discovery of homo naledi: even though specimens have, to the best of my knowledge, only been found in the one cave, they are nevertheless clearly of interest and scientifically significant. On the opposite end of the scale, if you want to assert that you have a better method of determining significant websites than PageRank, then it will likely be difficult to convince readers unless you apply to datasets of a similar scale to the "whole internet" scale that PageRank is routinely applied to.

Without knowing details about your exact image problem, it is difficult to know how much can be learned from a single image. In general, image processing is a very well developed field in which most things are applied to large data sets, so my zero-knowledge guess would be that you need more data (though since you say the prior papers have <10 images, it may otherwise). On the other hand, you might be doing something where four images are actually significant. Finally, you can always send it in anyway and see what feedback you get.

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .