Single image enlargement based on kernel estimation and linear weighting
I Komang Somawirata, Keiichi Uchimura, Gou Koutaki · 2013
This paper proposes a method for single image enlargement with linear weighting techniques and kernel estimation. The aims of our technique are to reduce the distance of the pixel value too far especially for interpolation pixel. Contribution from the closest pixels to the interpolation point is unchanged meanwhile the farthest pixel contribution will be estimated. There are four pixel contributions as a determinant of the interpolation pixel value. The four pixels are placed in a 2×2 kernel matrix. Each pixel in the kernel has a weighting value. The weight value is organized in the separate places that are placed on the weighting matrix with 2×2 sizes. Weight values obtained from the linear curve based on the position of the pixel interpolation. Improving the image quality is performed only on the interpolation pixels. Experimental results show our new method can produce a better enlargement result, especially in the edge image regions compared to the comparison methods that used in this paper.