Comparative Analysis in Pre‑image Algorithms of Kernel PCA

Canran Ji, Wojciech Czaja

Track: Full Paper (8 pages)

Keywords: kPCA, GAN, Neural Network, kernel method, DCGAN, WGAN, denoising.

TL;DR: We benchmark classical kPCA pre-image methods against two GANs and find that WGAN-KPCAnet delivers the most robust, high-fidelity reconstructions.

Abstract: We study the kPCA pre-image problem in image denoising by benchmarking classical algorithms and introducing two neural adversarial pre-imaging models, DCGAN-KPCAnet and WGAN-KPCAnet. Our results show that WGAN-KPCAnet delivers superior reconstruction results and is robust to noise compared to baselines.

Submission Number: 44