Linear interpolation with edge-preserving adaptive weights

To solve the problem of edge blurring and inaccuracy in image interpolation, we propose an edge-preserving interpolation method based on in-pixel grayscale modeling and adaptive weighted linear interpolation. First, edges are determined by integrating the grayscale changes of the image window and calculating the maximal eigenvalue of the result matrix. Then the eigenvalue and corresponding eigenvector are used to model the grayscale distribution inside the edge pixel. The adaptive weights are determined both by the Euclidean distance of the interpolation position to the neighboring pixels and the edge strength of the neighboring pixels, where smooth pixels are given larger weights. Finally, the interpolated pixel value is obtained by weighted linear summation of the neighborhood pixels. Experimental results show that the proposed method effectively preserves edges, suppresses common interpolation blurring and zigzags, and produces better PSNR and MPSNR on test images.