Rawsie
vs. other compression algorithms

Christoph Clausen · Dotphoton/Rawsie Chief Scientist · December 5, 2019

As any new technology, Rawsie might be tricky to understand. One of the top questions we get every day is "How does Rawsie optimization compare to other raw compression tools?". Here's a quick guide to help sort that out.

First of all, our average compression ratio is higher than anyone's on the market. Only lossy DNG (available in, e.g., Adobe Lightroom) is able to achieve something comparable. But, obviously, standard lossy compression hurts your image quality. Here's a chart showcasing file size ratios for Rawsie and other popular solutions compared to an unprocessed raw file as a 100% file size reference.

Sony compression: 50%

Lossless DNG: 50%

Nikon lossy NEF: 45%

Canon lossy CRAW (CR3): 45%

Canon SRAW, MRAW: 25%

Lossy DNG: 15%

Rawsie: 15%

Unproccessed raw: 100%

Canon lossless compressed: 50%

Nikon lossless compressed: 50%

As you can see, the smallest file size is achieved by lossy DNG and Rawsie. But in terms of quality the two are significantly different: the quality of Rawsie-generated files is much closer to the one of lossless DNG, at about 1/3 of the file size. Possible quality issues with lossy DNG and other formats are shown in the chart below.

Secondly, Rawsie is the first raw compression algorithm designed for both human eye and machine vision from the ground up. When we talk about image quality, it's not just a matter of visual taste and tests for us, or say, capability of any photographer to distinguish or not some possible artefacts. Rawsie only skips the information attributed to the randomness of photons, and that's why even professional software will not see the difference.

Existing raw image compression algorithms basically fall into two groups:

1

Camera based compression
This means your images are processed directly on your camera. Apart from obvious storage economy, this may also be caused by your camera's limited performance: for example, in some cameras' burst mode, manufacturers automatically downgrade your image quality to allow the camera to record the image quickly enough.

Camera based compression also has the disadvantage that you cannot decide at a later stage to apply compression. The image is either compressed when taken, or never.
2

Software compression
The most commonly used tools for raw image compression are Adobe DNG Converter and the raw export of Adobe Lightroom. Here, one can choose between lossy and lossless compression. As we've mentioned earlier, lossy will affect the image quality, while lossless will not, but will only provide about 50% file size reduction.

Now, let's put file size aside. As any artist cares for image quality, it's also our top priority. After all, none of us love the idea of spending thousands of dollars on amazing cameras just to get half of the quality we are striving for. Getting a top camera in any price range and then having to live with artefacts and unstable quality is something like – well, let's not go there...

As you can see in the list provided below, most of the algorithms and tools will somehow downgrade the image quality and may even introduce sometimes severe, irreversible artefacts:

Sony compression

50%

Posterization, visible typically as horizontal bleeding of high-contrast vertical edges appears.
The raw image pixel values are mapped to 11 bit. Pixels are 1-dimensionally grouped into chunks of 32 pixels, which are then quantized to an effective 8 bits per pixel. You lose some image information, in particular close to high-contrast edges. Posterization artefacts appear.

Lightroom-generated lossless DNG

50%

Basically nothing happens: all the original data is still there.
Some photography software may produce slightly different colors compared to files in the native camera format.

Nikon lossy NEF

45%

Raw pixel values are mapped to a bit depth that is about 2 bits lower than the original. No artefacts.

Canon lossy CRAW (CR3)

45%

Introduction of structure in chroma noise when trying to compensate in software for significant underexposure.
Algorithm is not really known. An initial guess is that wavelet transform with weak losses is applied. Noise pattern artefacts may appear.

Canon SRAW/MRAW

25%

Reduced image dimensions. Aliasing artefacts.
Pixel count is reduced by a factor of 4 for SRAW. MRAW is similar, but only reduces pixel count by a factor of 2. Color values are converted into a different color space, roughly separating brightness from actual color, and color information for every second pixel discarded. You lose pixel count, some color information.

Lossy DNG

15%

The raw image color range is mapped to 8 bit and lossy JPEG compression applied. You lose part of the original data from the color filter array of the camera. Strong edits of colors and contrast may reveal artefacts such as color banding and blocking.

Rawsie

15% of original size

Most of the random quantum noise of the image is processed. No pixel count loss. No artefacts, the final image may appear slightly sharpened for some camera models.

To sum it up, this balance of image quality and file size is what makes Rawsie stand out from the crowd. Image storage is a sensitive topic for any artist, but we'd love to encourage you to give Rawsie a try, and this is why we provide a limited free edition of Rawsie, where you can optimize up to 30 images a day for no charge to see it with your own eyes, or you can choose one of the paid plans depending on your payment preferences:

Cover image by Mukuko Studio

Subscribe to monthly Rawsie updates