This technology is a three-stage demosaicing method that performs weighted interpolation of missing green elements in eight directions—including vertical, horizontal, and diagonal—within Bayer pattern color filter array images. It then fills in red and blue elements based on their correlation with the restored green plane, followed by color-difference-based post-processing to refine all three color planes.
Conventional Bayer demosaicing primarily estimates green values using color differences in only two directions (horizontal and vertical), leading to reduced interpolation accuracy for diagonal edges or slanted patterns. Furthermore, it often fails to suppress aliasing noise remaining after interpolation, resulting in persistent color bleeding and moiré artifacts.
This technology applies an 8-tap interpolation filter to a 9×9 Bayer block to obtain green estimates in diagonal directions, then normalizes and sums the 8-directional estimates using weights derived from directional gradients. Subsequently, it uses a 5×5 sliding window with a 5-tap anti-aliasing FIR filter and inverse gradient weighting to refine red-green and blue-green color differences, thereby reducing interpolation noise. Applicable to image signal processors in digital cameras and smartphones, it enhances color reproduction in scenes with significant diagonal components, such as diagonal structures or fabric patterns.
N/A
