Ask Your Question
0

How do I determine redundant 2d data sources the best way?

asked 2017-12-25 04:10:19 -0600

Sylvia gravatar image

I am struggling with the following problem. Images are captured using different settings. Each images is used to generate a binary mask. All these binary mask are combined using the OR-operator in order to get the final binary mask. Multiple of the input masks might be redundant. In order to speed up the capturing process, I would like to determine the redundant binary masks. The tricky part is that possibly a the partial masks might contain single pixels that is not redundant, but I want to ignore this right at the moment. I have a data set of approximately 200 final masks and all the corresponding partial masks are available.

I tried to visualize the task and I would like to know what would be the best approach to solve this problem. Is is machine learning or is there an other solution?

image description

Thanks in advance for answers and inspirations

Sylvia

edit retag flag offensive close merge delete

1 answer

Sort by ยป oldest newest most voted
0

answered 2017-12-26 21:22:12 -0600

Tetragramm gravatar image

Well, a simple test is calculate the sum of the absolute difference between the masks. Then choose a threshold. If the difference between the masks is below the threshold, they are redundant.

For example, you might choose that at least 1% of the pixels must be different.

Alternatively, blur the difference image and see if all the differences are concentrated in one place. If it's just small differences, they should be scattered over the image, but if it's a real difference, it's likely in one place.

On the other hand, all of this only works if the cost of finding redundancies is less than the cost of just OR-ing them together. So if you can re-use this information over many sets of masks, that's good. ie: Mask 1 and 3 are always redundant. But if not, OR is very fast, much faster than this kind of processing.

edit flag offensive delete link more

Question Tools

1 follower

Stats

Asked: 2017-12-25 04:09:22 -0600

Seen: 164 times

Last updated: Dec 26 '17