Image enhancement application development I – enlargement and interpolation

Abstract

This project implemented a novel image interpolation algorithm called Image Mesh Data-Dependent Triangulation (IMDDT) for resolution enhancement comparison by using cross-platform Java language for implementation. It is a low-cost edge-directed interpolation method, and has been certified by experiments of its efficiency and high-quality results when comparing with other conventional interpolation schemes. Interpolation algorithms can be classified into different orders depending on their complexity. Nearest neighbour is zero-order interpolation and is a basic algorithm with low processing time but generates blocky image appearance. First-order interpolation like bilinear interpolation results in better images. It is more complicated as it considers the closest 4 pixels surrounding the interpolated pixel. Nevertheless, images may still be blurred. For higher-order interpolation, bicubic interpolation produces sharper images, but is much more computationally expensive. IMDDT, first-order with less computational intensity, has a higher efficiency than bicubic interpolation yet produces comparable quality. IMDDT is implemented by first determining the diagonal edge directions of 4-pixel squares. After that, inverse mapping is adopted to interpolate from the 3 vertices of triangles separated by the edges. IMDDT is a fast and simple approach with pleasing quality. Future developments include various image transformations, graphics card acceleration and other possibilities.