HDR experiences

[Dominique Galland]

In a derived project I have done some experiences with HDR.

Below an image that shows what brings the fusion of three images. 

The first is underexposed, the second overexposed, the third at the automatic exposure of the camera the fourth a Mertens merge and the last an HDR treatment.

We first see the great interest of this process.

More precisely, the merge Mertens brightens the dark and softens the whites but makes a scene unnatural.

A true HDR algorithm requires knowing the exposure time, it tries to reconstruct what would see a human eye by correcting the defects of the sensor.

Here I used a Merge Debevec followed by a tonemap Durand (I noticed that it was not more CPU consumer than Mertens).

The last image lacks a little brightness but I think it shows the quality of the 8mm film. t could be improved in a post treatment

All these algorithms are in OpenCV we can refer to

https://www.learnopencv.com/high-dynamic-range-hdr-imaging-using-opencv-cpp…
and
https://www.learnopencv.com/exposure-fusion-using-opencv-cpp-python/

The problem is also to choose the times of exposure,

it really needs a dark image and a clear image.

 In my case for this image the auto exposure of the camera was 0.14 ms, the dark time a factor 0.1 so 0.014 ms and the clear time a factor 5 so 0.7ms.

Regards

[Dominique Galland]

I post the result of another experience.

The same picture with 25 exposures from the minimum to the maximum and the last the result of an HDR treatment

It can be seen that none of the 25 images is really correct. The raspberry camera shows its limitations. The processed image is much better.

A merge or HDR treatment is really essential.