I am having hazy underwater image (As shown in Figure) and I want to estimate the depth information of the scene from a single image. What could be a better method for this?
Interesting Q a conjecture
1 Manually select a ref pixel point on surface water which is brightest in Image
matrix
2 Assume intensity is proportional to depth
3 Subtract 1 from image matrix .. and normalize ..this
might correspond to relative depth
4 To convert 3 to absolute units we may have
to derive the Camera pose from image
Good Luck
Cheers
If you have only different intensities of pixels describing blue water; you can simply use the upper mentioned suggestion " Assume intensity is proportional to depth".
But for this given image as you have another part containing some plants and rocks with some complicated features. I think you need to use an already trained ANN (You don't have a sufficient number of data set to train it, in this case).
If you prefer to test classical solutions; I think you may get a promising result if you test Gabor filter + PCA/K-means; using this technique for this given image you may arrive to distinguish between three distinct regions (The rocky region + 2 regions of deep water and shallow water). In any case, it is hard.
May be for starters you can use an edge detection operator and visually compare !!
Narasim Ramesh Thanks for the answer. The scheme sounds interesting. Could you provide a reference to this work?
Deepak Kumar Rout You can go through the below paper by He et al.
Article Single Image Haze Removal Using Dark Channel Prior
This algorithm can extract depth information from a single image accurately.
Deepak I do not have a reference .. It was just a conjecture with stars in mind
assuming that only reflection takes place .. ignoring dispersion ,refraction etc..
the reflection intensity from objects would be inversely proportional to distance..
For example from astronomy see
https://courses.lumenlearning.com/towson-astronomy-2/chapter/the-brightness-of-stars/
However Subhadeep has given a much better ,more useful reference .
Good Luck
Cheers
- Badges
- Science topic
- Similar topics
- Computer Science
- Images