This technology is a demosaicing method that calculates an Optimal Bayesian Estimate (OBE) for Bayer pattern images from single-sensor cameras. It first reconstructs the green channel, then uses it as a reference to reconstruct and refine the red and blue channels. It employs a hybrid approach that weight-combines horizontal/vertical components with adjacent pixel patch similarities.
Single-sensor digital cameras use color filter arrays to reduce costs, meaning each pixel records only one color. Consequently, a demosaicing process to interpolate the remaining two colors is essential. Conventional interpolation methods often suffer from zipper artifacts at edges, color distortion, and significant image quality degradation in noisy environments.
By considering both point and path similarities through Bayesian estimation and iterative refinement of the green and red/blue channels, this technology improves CPSNR, S-CIELAB, and FSIM metrics while reducing zipper artifacts without sacrificing processing efficiency. It effectively handles images with Gaussian and Poisson noise. Applicable to ISPs in smartphone cameras and automotive/security sensors, it enables sharp color images even with low-cost sensors.
N/A
