Image / Video Subtraction techniques

[hhs_89]

I am new to Image processing and have learnt only few bits using Octave + net resources.

I am trying to subtract two images A & B. A contains noise and B doesn't. Both of these images are gray (blank shaded of gray white image) and the noise looks like a a few dark shaded pixels runnnig horizontally in image A which can be distinguished visually.

Using imsubtract method deosn't give me the result I want, so I am looking for advanced techniques / steps that can help this issue.

The noise on the image is due to flicker on LCD screen which is captured by camera. I also plan to apply this subtraction technique(s) on the video of under same conditions.

In tutorials I found people subtracting videos but those are rich in color, my input image is lacks color. But definitely running through pixel values will result in some differences.

Aany ideas? Am I thinking correctly about this problem by utlizing FFT and pixel subtraction based techniques?

[snibgo]

I'm not sure what you are trying to achieve. Perhaps: given a noisy and noiseless version of the same image, assume other images have the same noise, so de-noise them. Is that right?

I suggest you upload example images. You can upload to somewhere like dropbox.com and paste the URLs here.

[hhs_89]

Snibgo, here is the link two images taken at different instant of time which I am try to process:
https://www.dropbox.com/sh/nbzw0zl3gojn ... rqCba?dl=0

[hhs_89]

Basically I am trying to establish the fact that image captured at two different instants of time show differences, in other words they are not 100% identical. I am not sure how to represent & find the differences. Few things I can think of is brightness, lighting etc.

As you can see from the images that there visual difference can be seen, the "grey area" has changed it's position.

[snibgo]

Code: Select all

convert dsc_001.jpg dsc_002.jpg -compose Difference -composite -auto-level td.png

This shows variation in noise, and possibly also a displacement.

Opening the two images as layers in Gimp, difference mode, there seems to be a small movement and rotation: approximately modelled by a horizontal movement by 2 pixels at the top and 1 pixel in the opposite direction at the bottom.

If you have raw images (instead of JPEG), they might be more useful. Or they might show even more noise.

[hhs_89]

Thanks, I checked with Gimp and I can see the movement. I will try to take more images. I am using camera on phone it saves only jpeg format unless i find a way to save in raw format.

[snibgo]

If you can greatly increase the lighting, the noise should decrease. This would greatly help the comparison process. My script whatTrans.bat can't find the translation, because it is swamped by the variation in the noise.

[hhs_89]

I can control the lighting by adjusting LCD backlight control. However, if with more lighting flicker is hard to observe. So there is a specific light setting which I use to detect if flicker is present.

(I am using max ISO on camera which is 20 MP)