Development of a Multi-Constraint Loss Function for Image Recovery from Pruned Features
Abstract
Existing image recovery methods have demonstrated that original images can be reconstructed from full features. In this work, we consider a more challenging problem of recovering images from pruned features learned by deep neural networks. Our study addresses this issue by introducing a multi-constraint loss function that integrates L2 distance, sixth-power summation, and total variation regularization to enhance reconstruction quality. The loss function enhances image smoothness and fidelity while ensuring that reconstructed images are encoded as vectors closely aligned with the pruned feature.The proposed loss function enables robust image recovery, preserving key visual features even at high pruning ratios. Additionally, this study investigates the impact of different pruning levels on reconstruction fidelity, highlighting the trade-off between pruning efficiency and recoverability. These findings provide valuable insights into inverse problems in deep learning and image processing, with implications for security risk assessment, efficient model evaluation, and feature redundancy analysis.
Published
2025-07-02
Section
Regular articles
An author's submission implies that the manuscript has not been published previously, and is not currently submitted for publication elsewhere. Submission also implies that the Corresponding Author has consent of all authors (the Authors). Upon acceptance for publication transfer of copyright will be made to the Publisher of REV-JEC, who guarantees that full content of the published article is freely distributed on the Journal's website. The copyright transfer gives the Publisher of REV-JEC full authority to resolve any complaints of misuse or abuse (such as infringement or plagiarism) of the published article. The Authors have the freedom to redistribute and reuse the published article in any medium or format for any purpose, provided the original published article is properly cited. An article submission implies author agreement with this policy.
