This technology is a directional interpolation method for color filter array images in digital cameras. It identifies edge regions and determines the interpolation direction for a central pixel by examining horizontal and vertical gradients at red sampling positions alongside edge directions of surrounding blue sampling positions using majority voting. Based on the determined direction, it restores missing green components by leveraging cross-channel correlations.
Single-sensor digital still cameras record only one color per pixel, requiring the other two to be estimated from neighbors. While conventional color-difference-based demosaicing is efficient in flat areas, it often produces jagged edges and texture artifacts when interpolation directions are misestimated. Methods utilizing non-local redundancy also fail to fundamentally reduce directional misclassification in complex regions.
This technology suppresses directional errors by summing direction flags of the central pixel and four neighboring blue pixels, adopting a horizontal or vertical direction only when at least four out of five match; otherwise, it utilizes all four directions. It then refines green values using inverse gradient-weighted averages to preserve smooth edges and fine details. Applicable to ISP pipelines in digital cameras and modules, it produces sharp color images with reduced zipper artifacts, even in complex, high-texture scenes.
N/A
