Image interpolation by direct cross downsampling

Direct downsampling (DD) based interpolation system can be used as an encoder-decoder framework for secure data transmission, while DD is often used as a data compression way and interpolation is used for decompression. In this paper, we propose a new direct cross downsampling (DCD) algorithm, which has the same function as DD. Compared to DD, the DCD has three main differences from the angle of image interpolation: 1) DCD can use the nearest neighboring pixels for image interpolation to effectively reduce the interpolation error; 2) More pixels can be interpolated in the first step in those two-step interpolation methods with the help of the DCD; 3) The DCD preserves the information of each column or each row in the original high-resolution (HR) image while DD discards the even columns and rows. On the basis of the proposed DCD algorithm, we further propose four DCD-based interpolation methods: DCD-based bilinear interpolation, DCD-based bicubic interpolation, DCD-based content-adaptive interpolation (CAI) and DCD-based edge-oriented interpolation (EOI). By comparing the proposed four DCD-based image interpolation ways with those corresponding DD-based interpolation strategies in both experimental and analytical ways, we have proved the superiority of the DCD algorithm. Furthermore, we compared the proposed DCD-based EOI with three previous excellent DD-based interpolation works and experimental results demonstrate that the proposed DCD-based EOI has the best performance on PSNR and SSIM.(c) 2022 Elsevier Inc. All rights reserved.

Automated summary

This paper introduces a new direct cross downsampling (DCD) algorithm and its DCD-based interpolation methods. By comparing with the traditional direct downsampling (DD) algorithm, experimental results demonstrate the superiority of the DCD algorithm.