Arbitrary scaling of images using an M-channel DFT filter bank with optimized adaptive interpolation kernels

Abstract

Abstract:A new approach to arbitrary scaling of high quality images is proposed, whereby a 2-D M-channel DFT filter bank is designed along with its synthesis part being modified on the basis of a compactly supported sampling function. For that purpose, the 2-D down/upsampling formulae in a closed form are derived and utilized. Also, an optimized adaptive interpolation technique is employed to compensate for quality degradation arising in scaled images. Finally, simulation results demonstrate that high quality images of arbitrary size can be obtained from the original image.

Design of a new DFT filter bank with modified sampling kernels and its application to arbitrary image scaling

“…2 (b) and Fig. 2 (c) correspond to 'Barbara' images, reconstructed from the original image by the proposed 3-ch MDFT FB and by the conventional 3-ch DFT FB with optimized adaptive interpolation (OAI) kernels [3] (here, α = 1). We can see from Fig.…”

“…Furthermore, a CSSF of degree 2, ψ 3 [s],0 (t), with a finite support of [−2 ∼ 2] [5], is employed, instead of the sinc function [7], as a sampling kernel, since the sinc function has some problems in practical implementation (e.g., truncation effects) [3,8]. Accordingly, for the design of M-band low-pass prototype pre/post filters, the following down/up samplers with a factor L (L < M) can be utilized:…”

“…In this paper, we propose a new MDFT FB for arbitrary image scaling by modifying further the conventional MDFT FB in a more efficient manner. In particular, the proposed MDFT FB does not employ the OAI technique which was employed to compensate for the image quality degradation [3] for further reduction of computational complexity. More specifically, the proposed MDFT FB is designed by modifying the analysis and synthesis parts of a conventional MDFT FB [4] (i.e., see (5)- (6)) as presented in Fig.…”

“…Thus, both analysis and synthesis filters, satisfying perfect reconstruction conditions, should be redesigned. To solve those problems, an arbitrary image scaling method, utilizing a DFT FB with closed-form sampling kernels, was recently reported [3]. While the DFT FB has some advantages such as high-speed implementation, it is not suitable for subband coding, since it is not easy to eliminate aliasing effects due to subsampling [4].…”

“…While the DFT FB has some advantages such as high-speed implementation, it is not suitable for subband coding, since it is not easy to eliminate aliasing effects due to subsampling [4]. Also, another DFT filter bank designed by employing an optimized adaptive interpolation (OAI) [3,5] was proposed to compensate for such quality degradation in the image reconstruction. However, the OAI technique requires high computational burden [5,6].…”