Bayer Demosaicing Method Using Bilateral Filtering and Directional Differences

Low-Noise Bayer Demosaicing Based on Bilateral Filter Weights and 4-Directional Differences

This technology is a demosaicing method that fills in missing color components from raw Bayer CFA data in single-sensor cameras. It first interpolates green values at red and blue positions using 4-directional differences (north, south, west, east) and bilateral filter weights that evaluate spatial proximity and pixel similarity, then estimates the missing red and blue values based on these results.

Commercial digital still cameras use a Bayer filter over a single CCD, recording only one color per pixel, which requires the other two colors to be reconstructed from adjacent pixels. Conventional color interpolation methods often leave behind noise and unnatural color artifacts, making it difficult to obtain sharp images.

This technology utilizes horizontal and vertical differences derived from first- and second-order derivatives alongside bilateral filter weights, further refining the interpolated green values using weight parameters. It can be applied to image processing pipelines for entry-level digital cameras, webcams, and industrial machine vision cameras, enabling the restoration of clean, natural colors without requiring a separate noise reduction process.

‍

‍

Key Features:

• A step of calculating 4-directional differences (north, south, west, east) at red pixel positions in the Bayer CFA pattern and interpolating green pixel values
• An interpolation configuration that uses weights evaluating directional differences, spatial proximity, and pixel similarity, with the similarity evaluation window size set to 1
• A step of refining the interpolated green pixel values using a formula with weight parameters of 0.25 and 800
• A step of estimating missing red or blue values using Bayer pattern red/blue values and interpolated green values, and interpolating missing colors at green positions

‍