Aplikasi Inverse Backpropagation Pada Penskalaan Citra Menggunakan Bilinear Interpolation

Abstract

Bilinear interpolation is a method that is widely used in image scaling where bilinear interpolation can be applied to upscaling and downscaling processes. Several previous researches have shown that bilinear interpolation provides low quality reconstruction results compared to other scaling methods but is still a good alternative considering the lower process complexity compared to other scaling methods. For this reason, a mechanism or model is needed that can be juxtaposed with bilinear interpolation scaling so that the reconstruction results have better quality. Referring to previous research, neural networks can be used in the reconstruction process where artificial neural networks are used to learn features or information that is lost during the downscaling process so that it can be reused during the reconstruction or upscaling process. This research applies inverse backpropagation to help improve the quality of image reconstruction results on bilinear interpolation. The test results show a much better MSE value of up to 40.25% compared to reconstruction using ordinary bilinear interpolation. Meanwhile, the increase in PSNR obtained ranged from 0.4% - 9.7%.