I have been working for some time in several kinds of image recognition. They work well. One example is , character or shape recognition. I have implemented this in an embedded processor and I can recognize shapes and letters with this.

Now I am working on implementing this with a Hololens. Using my algorithms I have already achieved an automatic digitalization of images (meaning that I can convert a image taking by the camera to black and white using automatic thresholds).

My question is the following:

So far my work has been on controlled environments. Meaning that I pointed the embedded camera exactly to a piece of paper with the shapes or characters I wished to recognize.

Now I am working on a uncontrolled environment. The hololens takes photos of the general environment surrounding me. I wish to recognize shapes or letters there but they can be anywhere. The possibilities are more varied than before.

I would like to get advice on how to treat this, where can I find good resources, books tutorials etc to address this problem.


  • $\begingroup$ I am not sure if this the correct platform to ask such sort of question you could check out stat.stackexchange.com. also for uncontrolled environment first you could try segmenting your image into regions by using some sort of clustering algorithms .Such as K-means or K-medoid. $\endgroup$ Commented Dec 8, 2017 at 12:28
  • $\begingroup$ I don't understand what you are asking. Are you asking how to convert to black-and-white? Or are you asking how to recognize objects in the image? Or are you trying to do OCR of text that is visible in the image? Or something else? Please edit your question to be more specific about exactly what task you are trying to solve. Otherwise, you're close to asking for a summary of the entire field of computer vision, which is too broad for this site. $\endgroup$
    – D.W.
    Commented Dec 9, 2017 at 6:13
  • $\begingroup$ I already did the black and white conversion. In fact I have already programmed algorithms for labeling, automatic thresholding, fourier descriptors etc. All of these tried on controlled environments. I was worried how to apply this on normal environments and asking for people with experience on this $\endgroup$ Commented Feb 20, 2018 at 1:11


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.