This technology is a demosaicing method and apparatus that first restores the G plane from a Bayer image, creates pre-estimated values for the R or B planes using a guided filter with the restored G plane as a guide, and then calculates and interpolates the residuals from actual sample values to add to the pre-estimated values.
Digital imaging devices such as smartphones use a single sensor and a Bayer color filter array due to cost and space constraints. Conventional directional interpolation or non-local adaptive techniques require a separate refinement step after estimating the RGB planes to ensure quality, which leads to increased processing time due to this additional post-processing.
This technology uses a guided filter, which determines linear coefficients based on the mean and variance of the G plane within a local window, to pre-estimate the R and B planes while preserving edges. It then corrects errors by interpolating residuals obtained at original color component positions to G component and other color positions. Since residual correction replaces the need for refinement, it achieves excellent demosaicing results in a short time without separate post-processing. It can be applied to real-time previews and burst shooting paths in smartphone cameras, providing high-quality color images instantly while reducing processing latency.
N/A
