Kernel regression for image processing and reconstruction

A method of kernel regression and image interpolation for image processing, comprising:

• a. estimating local gradients of original data in an image using a classic kernel regression technique to provide local image structure data;
• b. analyzing said local image structure data and computing a scaling parameter, a rotation parameter and a elongation parameter by applying a singular value decomposition to said local gradient estimates to provide a steering kernel;
• c. applying a steering kernel regression having said steering kernel to said original data and having a reconstructed image outcome with new local gradients in said image;
• and
• d. iteratively applying said steps b and c to each new said reconstructed image outcome up to ten iterations, wherein said image has said gradients containing local image structure information and each said iteration improves said gradient and pixel values of said reconstructed image, and wherein said original data is denoised and said local image structure information in said reconstructed images is improved.