Better visibility with software: removing haze and colour cast from your drone footage

You can already gain a lot during the dive itself: lights down, lamps away from the camera, calm flying. This article picks up after that. What can you get out of the footage you already have with software?

Important to know up front: we have not yet compared the programs and filters named here using our own drone footage. Everything is based on manufacturer specifications and specialist sources. It is an assessment, not a hands-on test.

Why the picture turns green and flat

A quick reminder: water swallows light, and not all colours equally. On top of that come suspended particles, which scatter light and lay a veil of haze over the picture. What you see is therefore a mixture of subject, colour loss and scattered light. We explain this in detail in Colours underwater: why everything turns blue, and what helps.

Software tries to separate this mixture again. The more real image information is still in the footage, the better this works.

Level 1: what the drone apps can do

QYSEA FIFISH

For the FIFISH V-EVO, QYSEA names two functions:

• “Real-time Color Balance & Image Enhancements”: a dedicated, adaptive image system that, according to the manufacturer, is meant to restore colours “especially in dark and turbid environments”.
• “Adaptive AI Sea Plankton Filtering Algorithm”: According to QYSEA, the V-EVO uses it to filter out white flakes caused by plankton and bacteria.

The description of the FIFISH app also speaks in general terms of automatic algorithms for colour enhancement and manual settings.

Chasing

For the Gladius Mini S, a retailer names “color correction” as a feature. We found no more precise details from Chasing on how this correction works.

Our assessment: in-app filters are convenient, but they often act directly on the recording. If you are unsure, check in the app whether you can additionally save an unedited original. That keeps post-processing open to you.

Level 2: post-processing on computer and phone

DaVinci Resolve (video)

DaVinci Resolve from Blackmagic Design is a video editing and colour correction program. According to the manufacturer, the free version is “packed with more features than most paid software applications”. The paid Studio version costs US$295 on the US site (no guarantee, as of October 2026) and, according to Blackmagic, brings temporal and AI-assisted noise reduction, among other things. Noise in particular is an issue with dark lake footage.

Typical workflow for underwater video:

1. Set the white balance on something that should actually be neutral (a light stone, sand, a grey card).
2. Pull back the green or blue cast using the colour wheels.
3. Raise contrast carefully, lower the blacks a little.
4. If needed, noise reduction (in the Studio version).

Adobe Lightroom (photos and stills)

In its own guide to underwater image editing, Adobe suggests these steps:

• White balance with the eyedropper on a neutral area or via the temperature and tint sliders
• Colour mixer to adjust individual colours precisely
• Masks (also AI-generated) to edit subject and background separately
• the Texture, Clarity and Dehaze sliders for more detail
• the Remove tool to stamp out individual particles

The “Dehaze” slider is designed for exactly the veil that suspended particles create. Caution: too much of it amplifies noise and makes colours tip over.

Dive+ (phone app)

Dive+ advertises in the App Store that it restores the true colours of underwater photos and videos “with just one tap” and for free. There are in-app purchases for extra features. In a comparison, the diving magazine Divernet praised the one-tap correction as often effective, but sometimes fooled by certain colour palettes and limited manually. The test dates from 2020, however.

Level 3: what research can do

Behind many filters are methods from image processing. You should know these four:

Contrast equalisation (CLAHE)

CLAHE stands for “Contrast Limited Adaptive Histogram Equalization”. According to the OpenCV documentation, the method divides the image into small tiles, equalises the contrast in each tile separately and limits it so that noise is not amplified too much. It is fast enough for live video and is built into many tools.

Dehazing (haze removal)

A well-known method is the “Dark Channel Prior” by He, Sun and Tang (2009). The idea: in clear outdoor shots, most small image patches have at least one very dark colour channel. If this darkness is missing, haze is involved and can be calculated out. The method was developed for landscape photos, but the idea is also used underwater.

Sea-thru

Sea-thru by Derya Akkaynak and Tali Treibitz (2019) physically calculates the water out of the image. The catch: Sea-thru needs a distance map for each image. The researchers generated these by photogrammetry from many photos and placed objects of known size in the scene for scale. For a single drone video without depth information, the method is therefore not directly usable. (More on the technology behind it in our article Photogrammetry: 3D from the lake bed (planned).)

One approach that aims to apply Sea-thru live to drone video is the Visual Recovery project, which we will present in a profile of its own (planned).

AI methods

Neural networks learn from pairs of “bad” and “good” images what an enhanced underwater image looks like. Two well-known examples:

• UIEB and Water-Net (2019): a dataset of 950 real underwater images. The authors criticise that many methods until then had only been tested on synthetic data or a few selected photos.
• FUnIE-GAN (Islam et al.): a network that, according to the authors, is fast enough for real-time pre-processing on underwater robots.

And how do you measure whether an image is “better”? Research is being done on that too, for example with assessment methods that need no reference image, such as NR-UWIQA (Irshad et al., 2021).

Our assessment: AI can deliver impressive results, but it can also “invent” details that were not there. For nice pictures, that does not matter. If you want to know whether there is really an object on the bottom, caution is advised.

The limits: what software cannot rescue

• No information, no result: what gets lost in noise or in the bright wall of flakes, no filter brings back.
• Blown-out areas: areas overexposed by lamps that are too bright are pure white. There is nothing left in them.
• Heavy compression: live transmissions and heavily compressed videos have less in reserve than the recording on the drone's memory card.
• Motion blur: fast runs in the dark produce smeared pictures. Sharpening does not turn them into a sharp picture.
• Stirred-up mud: when visibility is zero, the only help is waiting or a sonar (see our article Adding sonar: what does it bring to an underwater drone?).

That is why the rule is: the best software is good source material. Lighting, distance and flying style at the dive site usually bring more than any slider afterwards.

Overview: which tool for what?

Prices without guarantee, as of October 2026. US dollar price from the manufacturer's US site, plus taxes customary in your country.

Practical tip to finish

Before the dive, film a white or grey card for a few seconds at the depth where you will be operating later. In post-processing, that gives you a reliable reference point for white balance, instead of having to guess which stone is actually grey.

Frequently asked questions

In video

An example of colour correction in DaVinci Resolve:

Lost something important underwater that a hobby drone can no longer recover? UDRA offers professional recovery missions in Austria.

EditorKennyKenny writes about ROVs, tests and missions for Unterwasserpilot.