Title:	Improving Image Reconstruction using Incremental PCA-Embedded Convolutional Variational Auto-Encoder
Authors:	Azizi, Amir Charambous, Panayiotis Chrysanthou, Yiorgos
Citation:	WSCG 2024: full papers proceedings: 32. International Conference in Central Europe on Computer Graphics, Visualization and Computer Vision, p. 107-116.
Issue Date:	2024
Publisher:	Václav Skala - UNION Agency
Document type:	conferenceObject
URI:	http://hdl.handle.net/11025/57383
ISSN:	2464–4625 (online) 2464–4617 (print)
Keywords:	zpracování obrazu;rekonstrukce obrazu;analýza hlavních komponent;konvoluční variační automatické kodéry
Keywords in different language:	image processing;image reconstruction;principal component analysis;Convolutional Variational Auto-encoders
Abstract in different language:	Traditional image reconstruction methods often face challenges like noise, artifacts, and blurriness, requiring handcrafted algorithms for effective resolution. In contrast, deep learning techniques, notably Convolutional Neural Networks (CNNs) and Variational Autoencoders (VAEs), present more robust alternatives. This paper presents a novel and efficient approach for image reconstruction employing Convolutional Variational Autoen coders (CVAEs). We use Incremental Principal Component Analysis (IPCA) to enhance efficiency by discerning and capturing significant features within the latent space. This model is integrated into both the encoder and sampling stages of CVAEs, refining their capability to generate high-fidelity images. Our incremental strategy mitigates scalability issues associated with traditional PCA while preserving the model’s aptitude for identifying crucial image features. Experimental validation utilizing the MNIST dataset showcases noteworthy reductions in processing time and enhancements in image quality, underscoring the efficacy and potential applicability of our model for large-scale image generation tasks.
Appears in Collections:	WSCG 2024: Full Papers Proceedings

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.