Demosaic JPEG Software

Updated: July 30, 2026

Demosaicing is a transformation of Bayer image into the conventional 24/48-bit RGB format. Demosaicing is required because digital cameras usually don’t produce ready-to-go RGB images, instead they store visual information as a set of separate R, B and G pixels derived from the image sensor of the camera, and the actual color of a pixel in that set is determined by interpolating nearby pixel colors. To high quality demosaicing algorithm we have added JPEG compression to fulfill the following:

• High quality image demosaicing on NVIDIA GPU, much faster than on any CPU
• Fast JPEG compression on GPU

What's unique in Demosaic JPEG Software

• This is the fastest demosaic and jpeg compression tool
• Sophisticated demosaicing algorithms with high PSNR and low artifacts
• Simple to use, fast to work and easy to check
• Created by company with years of CUDA experience and world-class achievements in the field

GPU JPEG Demosaicing Software Features

• All CFA Bayer mosaic patterns for input data supported (RGGB, BGGR, GBRG, GRBG)
• Read 8/16-bit input CFA image data from HDD/SSD/RAID in PGM format or from PC memory
• Output format: 24/48-bit JPEG with variable compression ratio
• Minimum image resolution 128×128 pixels
• Maximum image resolution up to 16,000×16,000 pixels and more
• Demosaic+JPEG performance without host-to-device and device-to-host transfers) on NVIDIA GeForce GTX 1080:
  • 2 MPix 8-bit image (DFPD demosaic, jpeg quality 90%, 4:4:4) – 0.57 ms
  • 8 MPix 8-bit image (DFPD demosaic, jpeg quality 90%, 4:4:4) – 1.7 ms
• Strong color moire artifacts suppression
• Timing and performance measurements
• Command-line application for Windows-10/11, Linux, L4T

Quality for combined Demosaic and JPEG

Both demosaicing and JPEG compression lead to image quality degradation. That's why if we apply these algorithms to image processing pipeline we could get significant quality loss for final image. It happens quite often because high quality demosaicing algorithms for CPU are very slow, so the choice of fast and simple demosaicing algorithm is usually straghtforward. But this is the way to get maximum image degradation.

In tables below we can see averaged PSNR/SSIM results for Kodak image set (total 24 images) which is widely used for quality evaluation in demosaicing. For testing we used standard Baseline JPEG algorithm and Bilinear/HQLI/AHD/DFPD demosaicing algorithms.

Demosaic and JPEG with subsampling 4:4:4

Demosaic and JPEG with subsampling 4:2:2

Demosaic and JPEG with subsampling 4:2:0

Now it's obvious to evaluate image degradation for combined JPEG+demosaic solution and make a right choice for both demosaic algorithm and JPEG quality. This is just an idea how to choose image processing algorithms and their parameters. Full solution should take into account compression ratio, total performance benchmarks, hardware pricing, and much more.

Bilinear demosaicing algorithm has good speed, though quality is moderate according to PSNR and SSIM metrics. Quality loss after bilinear algorithm is significant and JPEG compression doesn't really change PSNR/SSIM after demosaicing. That's why we would suggest to choose demosaicing from HQLI/AHD/DFPD or other high quality algorithms.

Demosaic JPEG FAQ

Can you demosaic and JPEG-encode in a single GPU pass?

Yes. Bayer raw is debayered to RGB and JPEG-encoded in one pass on the GPU, without a round trip to CPU memory between the steps — which is what keeps the pipeline fast.

Which JPEG subsampling modes are supported?

4:4:4, 4:2:2 and 4:2:0. Use 4:4:4 for maximum color fidelity and 4:2:0 for the smallest files; the demosaic step feeds the JPEG encoder directly in each case.

What else is interesting in Fastvideo GPU solutions

We are also developing fast image processing software on GPU for high speed and high resolution cameras in real-time: dark frame subtraction, flat field correction, white balance, image filtering, noise removal, color correction, demosaicing, jpeg/j2k compression, etc.

About the author

Fyodor Serzhenko, PhD, is the founder and CEO of Fastvideo. He earned his PhD at the Moscow Institute of Physics and Technology (MIPT) in 1993. Since 2009 he has led the development of Fastvideo’s GPU-accelerated image codecs and ISP modules, including the Fastvideo SDK. Connect on LinkedIn.

Why you can trust these results

Fastvideo has built GPU-accelerated image processing software since 2009. Our codecs comply with the standards they implement — JPEG (ITU-T T.81 / ISO IEC 10918) and JPEG2000 (ITU-T T.800 / ISO IEC 15444) — and our tools are open source on GitHub. Every performance figure we publish is reproducible: see our benchmark methodology, download the Fastvideo SDK benchmark report (PDF), and measure it on your own GPU.