Adaptive multidirectional edge directed interpolation for selected edge regions

This paper proposes an adaptive edge-directed interpolation algorithm using multidirectional neighbor pixels. In order to restore multidirectional edges, a missing pixel is estimated as a weighted sum of 12 neighbor pixels. Based on the geometric duality between a low resolution image and a high resolution image, interpolation coefficients are predicted using Wiener filter theory. In order to reduce the computational complexity, interpolation region selection method is also proposed. An edge map for a low resolution image is obtained by canny edge detector. By analyzing edge continuities, only long edge regions are interpolated using 12 neighbor pixels. Short edge regions are interpolated by new edge-directed interpolation, and even regions are interpolated by a linear interpolation. Simulation results show that a proposed method restores major edges with several directions better than other methods in subjective tests while showing fair performance in objective tests.

I. Introduction

Image interpolation is a process of generating a high resolution image from a low resolution image, which is one of the key issues in image processing field. Various applications such as demosaicking, image enlargement and surveillance camera image enhancement require high performance interpolation algorithms. For preserving valuable details of a low resolution image without artifacts, an interpolation algorithm has to reflect image features carefully and considerably.