8

I've been playing around with making images better and removing all the unwanted disturbances from it. I found out a major concern would be removing the unwanted reflections that might come.

I've gone through a paper that uses ICA to get rid of these reflections. What other methods would work?

Say I have an image:

enter image description here

Another image:

enter image description here

Converted it into CMYK

enter image description here

How should i remove this semi black portion?

user
  • 103
  • 5
vini
  • 2,182
  • 4
  • 21
  • 37
  • Note that the approach described in the paper you have linked to requires two images with a different polarization. Do you have similar requirements? 1/ Should this work automatically or is human supervision allowed? 2/ Should this work from a single image or is it OK to require the user to produce various images of the same scene? 3/ Should this work with any photographic image or is it OK to require the use of special filters/optical devices on the camera? – pichenettes Apr 07 '12 at 16:04
  • Well its not neccessary to use the same approach as they have proposed and i don't have two images just many images with reflections that i would like to make as less as possible – vini Apr 07 '12 at 16:12
  • Did you read other papers too? I get loads of relevant hits with the queries "blind image separation", "glare removal", and "reflection removal". – Emre Apr 07 '12 at 20:12
  • Yes i found all of it most techniques are really complex and would take time for understanding what if this could be solved with easier techniques like filtering – vini Apr 08 '12 at 05:17
  • 3
    Your problem is complex; deceptively so. I do not think there is an easy way out. It's a testament to the ability of our brain that it seems otherwise. – Emre Apr 08 '12 at 18:48
  • there is always a way of finding such methods :) i guess ill embark on that journey – vini Apr 09 '12 at 04:45
  • Do you have multiple frames of motion? I don't think there's any way to identify the reflection from a single static image. The reflection information and other information has to be separable somehow. – endolith Apr 09 '12 at 14:44
  • maybe i can split the image into different planes as in convert it into 3D or something would that work? – vini Apr 09 '12 at 16:04
  • @vini: Do you have any information other than a single static image? – endolith Apr 09 '12 at 18:11
  • i have many such images ...mostly with a front view and sum having taken from behind a window – vini Apr 11 '12 at 10:37
  • @vini: Multiple images of the same scene? That might make it possible. Can you post examples? – endolith Apr 11 '12 at 16:59
  • I was looking for something else and saw your question as a related of what I was looking for. I'm not an expert in this area either, but Sing Bing Kang et al. from MSFT Research did some studies a long time ago about reflections and the results were quite good: http://research.microsoft.com/apps/pubs/default.aspx?id=67274 Though this is not an answer that will solve your problem, I just wanted to share with you the link just in case it points you in the correct direction. – SRG Jun 23 '15 at 21:40

1 Answers1

6

I am not a specialist in image processing but I highly doubt this is feasible with your requirements (single image, no prior information). The challenges is twofold:

  • Detect which area of the image is a reflection. This looks like a quite complex scene understanding problem. In your example, I know that the right side is a reflection because I can infer from the image that a person is depicted here, and I have prior knowledge about what the back of a person looks like.
  • Reconstruct the background in the identified region. While it would be possible to just inpaint the area detected as a reflection with textures from the surrounding, the result would hide many details! Whatever I've seen in the field of single input image separation was relying on very simple image models (stationary textures), not real world images. The more traditional image separation setups (multiple channels, recto/verso of translucent pages, multiple polarization) do not apply to your case.
pichenettes
  • 19,413
  • 1
  • 50
  • 69
  • 1
    what are the tradition image separation setups that you referring to ? – vini Apr 07 '12 at 19:27
  • 1
    One example is when scanning recto/verso documents. The text on one side of the page "bleeds" through the other side (http://ars.sciencedirect.com/content/image/1-s2.0-S0031320311003049-gr6.jpg). This is close to a traditional BSS setup since you have 2 sources, and 2 observations of the same sources with different mixture weights. Another examples are in multi-channel image analysis in astrophysics. These situations where you have as many images as sources to extract are the most commonly explored in the image separation literature. – pichenettes Apr 07 '12 at 21:12