Image interpolation device, method using modified kernel regression, and computer-readable recording medium storing the method

Image interpolation device with reduced edge blurring using smoothing parameter-weighted kernel regression

This technology is an image interpolation device that downsamples and upsamples images from color filter array sensors, then fills in empty pixels by calculating luminance values using a modified kernel regression that weights differences in pixel luminance and spatial distance.

Simple filter-based interpolation methods, such as bilinear and bicubic, often suffer from blurring and aliasing artifacts at image edges. Since human vision is sensitive to boundary information, these defects significantly degrade the perceived quality of enlarged images.

This technology adjusts the smoothing parameter h within a range of 1 to 11 to determine the bandwidth of the weights, controlling the smoothness of the kernel regression results to restore high-resolution images closer to the original. It can be applied to mobile device screens of various resolutions, as well as satellite and medical image viewers, providing sharp images where outlines remain crisp even when low-resolution sources are enlarged.

‍

‍

Key Features:

• An image sensor unit that receives color images via a color filter array and a sampling filter that identifies the position and luminance values of specific row and column pixels
• A processing unit that creates a low-resolution image by downsampling the sampled image, then upsamples it and interleaves it to fill empty pixels with zeros
• A channel interpolation unit that calculates empty pixel luminance values using a modified kernel regression that applies weight information based on pixel luminance and spatial distance between two pixels
• A configuration that adjusts the weighting range so that the estimated kernel regression result becomes smoother as the smoothing parameter h, set between 1 and 11, increases

‍