This technology is a demosaicing method that estimates green pixel values in four directions—north, south, east, and west—using a second-order Taylor series approximation. It interpolates missing green values by calculating a weighted average using adaptive weights, then refines the green values and estimates the missing red and blue values. The second derivative of the green channel is calculated as the average of the second derivatives of the red and blue channels.
Cameras using a single image sensor and a color filter array capture only one color per pixel, necessitating the interpolation of missing colors. Conventional methods typically detect edges first and select one of a limited number of candidate directions for interpolation, which often leads to color reconstruction errors if the wrong direction is chosen.
Instead of selecting a single direction, this technology combines estimates from all four directions using color-difference-based adaptive weights to avoid incorrect directional choices, and further refines the results using color differences derived from the reconstructed green channel. With low computational overhead, it offers high processing efficiency and improved image quality. It can be applied to built-in ISP chips in image sensors and low-power camera modules, enabling the production of full-color images with minimal color bleeding even in resource-constrained environments.
N/A
