DEMOSAIC

Digital cameras have an image sensor overlaid with a mosaic of color filters which sample a single color per pixel. A reconstruction step is required to output the color image which has three color components per pixel, from an input image having only one.

The core of our demosaicing method is to consider a neighborhood of pixels in the RAW image from the sensor and calculate the appropriate weights between these pixels to best reconstruct the red, green and blue pixels of the color image. We build a physical model of the image formation on the sensor that we use to learn the inverse solution.

Innovations in technology are:

• Generic method that works for any sensor
• Better pixel exploitation in terms of edge stitching and color accuracy even in low light
• Reduced calculation time compared to other methods
• Gain in consumption compared to competing methods

We have a demonstrator in Matlab in the form of a Toolbox which has enabled us to compare the technology with other methods for improving quality (resolution, DeltaE, PSNR) and processing speed.

The products and services envisaged are:

• RAW converter software to reconstruct color images from the RAW image
• Filter transmission calculation software
• Demosaicing bookstore