Image Autoregressive Interpolation Device and Method

High-Definition Upscaling Method Based on Two-Stage Diagonal and Vertical Autoregressive Interpolation

This technology is an image interpolation device and method that restores missing pixels using an autoregressive model with directional coefficients and Lagrangian coefficients, based on 8-connected and 4-connected neighbor pixel matrices around the target interpolation pixel in low-resolution images. It performs diagonal and vertical interpolation in two distinct stages.

Ultra HD products require real-time upscaling of low-resolution video, but conventional interpolation methods like bilinear, bicubic convolution, and bicubic spline often result in artifacts such as blurring, ringing, aliasing, and zipper effects. Adaptive interpolation, while offering better image quality, has high computational complexity, making real-time processing difficult.

This technology first interpolates dark points using diagonal directional coefficients, then treats the result as known pixels to interpolate gray points using vertical directional coefficients, preserving both edge sharpness and spatial detail. Since the two interpolation stages are pixel-independent, they are well-suited for GPU parallel processing. It can be applied to UHD TV/monitor upscalers, video scaling, and deinterlacing, enabling the near real-time generation of high-resolution images with minimal artifacts.

‍

‍

Key Features:

• Obtaining a low-resolution image by downsampling a high-resolution input image
• Generating an 8-connected neighbor pixel matrix and a 4-connected neighbor pixel matrix centered on the reference pixel to be interpolated
• Constructing a first matrix with diagonal directional coefficients and a second matrix with vertical directional coefficients using the two neighbor pixel matrices
• Using the reference pixel obtained from the first image interpolation as a known pixel for the second image interpolation to enhance output image sharpness

‍