I work in a research lab and am currently setting up a system to image objects approximately 30 microns in diameter with a 250x USB microscope, then use image subtraction to isolate the objects as they move past the microscope FOV, and use blob detection to detect the object's presence in the binary subtracted image.

The microscope can capture still images at 1280x720, but video can only go up to 640x480. Ideally I wanted to take rapid still images (~10 FPS) at 1280x720, but various issues are preventing me from doing that. Capturing rapid images at 640x480 on the other hand is easily doable, but I'm not sure if the images would be clear enough for the image processing to resolve the objects at that resolution.

Would capturing the images at 640x480, then resizing them to 1280x720 before processing them, improve the ability to detect/resolve the objects in the image?

  • $\begingroup$ How fast can you capture individual images rather than use a video format? That is your better option if the frame rate is high enough. $\endgroup$ Commented Dec 13, 2019 at 0:52
  • $\begingroup$ Could you please review my answer? If something missing let me know. Else, could you please mark it? $\endgroup$
    – Royi
    Commented Feb 6, 2023 at 9:00

3 Answers 3


It is theoretically impossible to add information by increasing the resolution of an existing image.

So from that point of view -- no.

However, I certainly find it conceivable that if you're using some canned algorithms they may be sensitive to pixelation, and such an algorithm may benefit from the image being upscaled and smoothed. But unless these unknown algorithms are really lame, I would only expect this to be an issue if your original images are very sharp -- if the un-differenced objects tend to have blurry edges, either because of the limitations of your microscope or the properties of the objects themselves, then I'd think about trying what you suggest.

Otherwise -- you can't get something for nothing!

  • $\begingroup$ Look again, it isn't quite a doubling, but I gave you an upvote anyway as this is nearly exactly what I would have said. $\endgroup$ Commented Dec 13, 2019 at 0:48
  • $\begingroup$ en.m.wikipedia.org/wiki/Data_processing_inequality. for more details $\endgroup$
    – user28715
    Commented Dec 13, 2019 at 0:50
  • $\begingroup$ The implicit assumption of your is not totally valid here. Increasing resolution doesn't add information in the meaning of the link but in the meaning of solving more data from given resolution it can (As data isn't vanished usually, just highly damped). $\endgroup$
    – Mark
    Commented Jun 16, 2021 at 16:24
  • $\begingroup$ That's the point of the whole paragraph on algorithms that are sensitive to pixellation. $\endgroup$
    – TimWescott
    Commented Jun 16, 2021 at 18:48

While indeed the Data Processing Inequality means you can't add information to your images by increasing resolution it doesn't say you can't transform your data into a form that gives you more data.

Moreover, your data isn't just images, it is a video.
So certainly given a video you can use modern methods to increase the resolution of the data to allow better performance from detection algorithms (Starting with the fast that the video can should even help you with denoising which of course will increase performance of the detectors).

So, use a modern method which both increase resolution and reduce noise of a video stream and you should get better results.

The only caveat is methods like that use some priors and assumptions on the data. Mostly about it being real world video while yours is from a microscope. You can either use them and hope it works or try something which was optimized for microscope.


Technically, yes. We did it before.

However, there's a twist. You need temporal data. Expand the picture to a bigger size was equivalent to obtain it to a bigger Fourier space, and Fourier space does not necessarily to obey the constrain by physical space. Thus, obviously, if the space domain was fixed, you need more time.

Image this, just like long time exposure can enhance the single on a single graph, if you got a bunch of graphs together in a short period of time, they will be able to compose and give you more detailed resolutions. However, there's several assumptions built in.

Otherwise, you need very specific mathematical constrains, and usually won't work very well.


Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.