how to get a uniform white balance on a set of images (timelapse)

Question

As I am realising a timelapse film, I've taken thousands of photos of a set and due to the different weather and light conditions the pictures are very differently exposed from sunshine to haze and rain.

I am looking for a way to generalise the white balance of all the images in order to have a timelapse as smooth as possible using python.

I already tried using defilcekr filter and correcting the white balance using a variable, but I couldn't find a way to generelise the parameters to all the images of different conditions.

here are some examples of the images I have :

did you **fix** the white balance and exposure settings of your camera while taking all those pictures? your camera could have applied arbitrary "artistic" filters that completely ruin any hope of getting this right afterwards. you need RAW pictures. -- and for this question to be successful you must post data. please review [ask] and [mre] — Christoph Rackwitz, Sep 08 '22 at 11:48
@ChristophRackwitz I just added some pictures to illustrate my problem. As for the camera I extract these images directly — coderAI, Sep 08 '22 at 12:35
I see now. there are a variety of things you can try, and see how you like them. -- pick a bit of grayish facade off the buildings in the back. use that for "white balance". quality of results depends on specifics of algorithm. and will be useless if the patch is overexposed. perhaps pick some of the concrete too... or give up and simply apply "gray world", or any number of other algorithms -- temporal blur ("ghosting"/motion blur). to even out the lighting changes... and the frantic buzzing about of people and the vibrating trees. — Christoph Rackwitz, Sep 08 '22 at 12:46

Christoph Rackwitz · Answer 1 · 2022-09-08T13:13:38.317

For a visually pleasing time lapse, there are many things you can try. I'd additionally recommend a temporal blur (motion blur) to even out local lighting changes and fast-moving objects. That also reproduces the impression of a normal video, which always has some non-zero exposure time, i.e. motion blur.

This code approximately works in a linear space, tries to balance colors (color temperature) and brightness. All this isn't physically accurate but close to what's doable with the information given. One glaring defect: I'm not "tone-mapping" here, so anything that's too bright is simply clipped to white.

The "Gray world" is just one possible way to estimate color balance. There are a bunch more, variously complex. Your original pictures may already be color-balanced okay...

def gamma_decompress(im):
    return np.power(im, 2.2)

def gamma_compress(im):
    return np.power(im, 1/2.2)

def measure_gray_world(im):
    avg = np.mean(im, axis=(0, 1))
    return avg

for im in images:
    im = gamma_decompress(im / 255)
    avg = measure_gray_world(im) # or im[height // 3:]
    im = im / avg * 0.3 # move mean down so picture isn't blown out
    im = gamma_compress(im)
    imshow(im)

You could also only respect certain areas of the picture for the color balance... say the lower two thirds, mostly ignoring the sky. That'll make the picture look colder because most of it is adjusted to be more neutral, and the sky gets to be very blue.

how to get a uniform white balance on a set of images (timelapse)

1 Answers1