Lecture 21: Image Sensors (74)
jsun28

I think it's really interesting how you can use SNR to calculate the clarity of an image. However, I'm not sure why we need to multiply the logarithm ratio by 20? Also if the noise level was infinitely large would this just produce a SNR of negative infinity? In this case, would the SNR essentially become indistinguishable when received in the real world?

Zzz212zzZ

The mean pixel value is essentially the photons that arrive at and are captured by our camera.

dhruvchowdhary

SNR is like a score that tells us how much clearer our image is from the unwanted fuzzy stuff, which is the noise. When we turn this into decibels, the 20 helps make the score easy to compare. If there's a ton of noise, the score gets low, showing the image is mostly fuzz and not clear.

Mehvix

@jsun28 this is just decibel convention

Boomaa23

Here we use a total distribution of data points to calculate SNR. However, is it possible to calculate SNR with only one data point? I would imagine so as at its core SNR is signal-to-noise ratio i.e. the signal vs the noise which could be generalized to any number of points. Is this a correct understanding of SNR in this context?

Edge7481

How does SNR tell us whether the image is noisy? How do you differentiate between images with lots of noise and images that simply just have lots of variety in colors and contrast

amritamo

In response to Edge7481, SNR can help us understand the balance between signal strength and noise in an image. The higher the SNR, the clearer the image with less noise. To discern between noisy images and those with variety in color and contrast, we can analyze noise randomness and spatial distribution. Noise appears random and evenly spread, leading to a flat frequency spectrum, while genuine image features display structured patterns aligned with the scene. Evaluating SNR alongside visual inspection helps distinguish noise from meaningful image content.

You must be enrolled in the course to comment