Request new feature AI powered Sensor Dust Spot Detector / Remover to remove those annoying and unwanted small sensor dust spots which spoil the ultimate quality of our images.
DXO provides many excellent tools to improve the quality of images and RAW images including to denoise and improve the lens optimisation of cameras. The one item which would round out these options is Sensor Dust Spot Detector / Remover. The feature would work as a two stage process to : a) identify sensor dust spots and then b) automatically heal or retouch the spots. Both parts of this process should have a granularity or strength adjustment settings so that the detection sensitivity can be set and the healing sensitivity adjusted.
With most modern DSLR/Mirrorless cameras small dust particles get inside the camera and even with manufacturers options to reduce dust or having your camera regularly cleaned professionally dust can accumulate on the camera sensor causing image imperfections. With DxO Denoising Technology and Lens Sharpening Optimization you get great image improvements - however these dust spots remain and can easily be hundreds per image.
Sensor Dust Spots are often present in all images, and most noticeable at 100% zoom or more common as you have cameras with higher resolution where on small piece of dust can affect a few adjacent pixels on the sensor.
Sensor Dust Spots are typically more visible in images shot with a high F stop value and will often appear as a round partially out of focus marks against the rest of the image. This is most noticeable when you have blue skies or other areas of reasonably uniform colour.
If you want to see how dusty your sensor is, take an manually-set out of focus image of a plain white or light coloured wall using a high aperture (F16 or higher) setting for your camera. Then open the image in your favourite editor and zoom in to 100% and look especially around the top of the image.
Sure sensor dust spots can be removed by many retouch tools, but that has to be done for every spot which can be many and then repeated on every single image. As an example some techniques which can be used in Photoshop use a combinations of min and max values plus thresholds which can help to create a layer mask. Once dust spots are identified, then existing retouch tools can be very effective at making corrections, but again this is a manual process for each image.
Hi, there are already some feature requests for something like this:
That isn’t to say that these proposals are the same as yours. We can vote for some or all of them. While an infallible process would use a blank photo as a reference for identifying dust spots, you seem to be suggesting that AI can be trained to recognize what doesn’t belong in the image and remove it intelligently. Is that so?
You created this topic as a feature request for PhotoLab, but didn’t set the sub-category to Feature Request. I’ve done that for you so that participants in the forum can vote for it if they want. I suggest you do the same by clicking on the Vote button at the top of the page.
1 Like
stuck
(Canon, PL7+FP7+VP3 on Win 10 + GTX 1050ti)
3
In which case you don’t need a dust removal feature, you do need to clean your sensor.
Yes there were some similar topics, but both mentioned sampling a set of images to detect dust - vs this request which is specifically asking to leverage the power of AI training to recognise what is a dust spot is in an image. (I believe this is something with Luminar Neo already do). As a human once you start to recognise these pesky dust spots you can quickly see them in many places by their telltale signs.
The biggest advantages of the AI approach is that: dust can easily move around on a digital camera sensor, even for images taken on the same day. If you just use average sampling across several images to create a dust mask, then this could quiet easily become irrelevant for the next bunch of images and even worse is that you would then be attempting to correct dust spots where none existed.
An AI approach to dust spot detection and removal would also work much better with different aperture settings, something which the sampling method would not work well for. This is because the dust spot on the sensor will have a different effect or be more detectable at high F-stop numbers compared to lower F-stops. So, if your dust spot sampling method happened across a bunch of images with F-stop between 5.6 and 11, then you attempted to apply this same masking to images taken in the same photo session which were taken with a F-stop of 16 or 22, then you would miss some of the dust spots and may undersize some of the spot detections as well.
An AI detection method, would typically find that camera sensor dust spots are always a local darkening of the image, as the dust is occluding some of the microlens elements of the sensor and reducing the light being recorded in that spot. Dust spots on the sensor are typically round fuzzy or out of focus darkening of the image a few pixels in size. Sometimes depending on the dust location relative to the microlens and which pixel colours they may be interfering with along with the way the sensor processes the image they can appear as two overlapping fuzzy dots. I am sure this could be something relatively easy to train an AI to recognise.
I have my sensors regularly cleaned often once a year, but dust spots can happen and build up quite quickly even if you are careful with your camera and have it cleaned regularly.
Even if you think your sensor is clean take some RAW images using a high F-stop preferrably F-16 or F-22 and you will often be surprised what you see, even a few days after haivng had the camera profesionally cleaned.
I rarely, if ever, physically clean my sensor. But then, I mainly use my 28-300mm lens, so it almost never leaves the camera body.
If I do change lenses, it is usually for a specialist shot with the 20mm or 85mm prime lens. If this becomes necessary in the field, I ensure that the front of the body points towards the floor and that the rear of the replacement lens is spotless and covered with its cap until the last moment.
With a bit of housekeeping, you really don’t need this kind of AI.
Yes, but some cameras, when new, tend to spray oil from mechanical shutter, which requires wet cleaning. My old D4 stopped “spitting” only after several months.
Also some lenses are notorious dust suckers (e.g. old Nikkor AF-S 300mm/f4 ED) but then using a good blower is usually enough.
I’ve never had it cleaned. I do as you do, point the camera face down in hand or on the tripod, have the replacement lens sitting upright on the loosened cap, so I only need to pick it up with one hand to mount it. The camera body is probably opened to air for two or three seconds. And my approach is to attach a lens and let it govern the way I shoot for a few weeks at a time. I do realize some have a need to frequently change lenses in the field, possibly in adverse conditions.
In response to all the Q, most of these talk about camera hygiene (see later) rather than the feature.
As to the feature request - of course, this should be done as an option, in exactly the same way as Denoise Technologies or any other feature in DxO PhotoLab. Then as mentioned in the request, it should have detection strength and correction strength adjustments so that as a user you can determine how well the AI model detects dust spots and how well it corrects them.
Regarding dust spots on sensors (camera hygiene) - I apply the same level of housekeeping and infrequently change lenses, but despite this and professional cleaning and taking precautions dust happens. If you have a mechanical shutter then there are moving parts inside the camera box which move air around. Since none of us change lenses in a clean room that air will always contain dust of some sort.
In the feature request, I mentioned you can observe dust sports when using high F-stop values. Even if you think your camera sensor is clean, try taking an image setting aperture priority so that you take the image using an F-stop of 16 or 22 (if your lens supports such values) and then open the raw image at 100% magnification You may be surprised to see dust spots, particularly at the top of your image if you are taking outside pictures, especially with blue skies present.