Document Type : Original Research Paper

Authors

• Mahsa Noohi ¹
• Javad Mohammadzadeh ²
• Mitra Mirzarezaee ¹

¹ Department of Computer Engineering, SR.C., Islamic Azad University, Tehran, Iran.

² Department of Computer Engineering, Ka.C., Islamic Azad University, Karaj, Iran.

Abstract

Background and Objectives: Content-aware image resizing via seam carving aims to adapt image dimensions while preserving salient and structurally important regions. Despite substantial progress, existing approaches still exhibit important limitations: deep learning–based methods typically learn saliency-oriented energy maps without explicitly enforcing structural continuity, whereas optimization-based seam selection strategies rely on handcrafted or shallow representations that lack semantic awareness. This mismatch may result in geometric distortions, such as structure bending or discontinuities, particularly in images with complex layouts. To address this issue, this study proposes a hybrid seam carving framework that integrates attention-based energy estimation with optimization-driven seam selection.
Methods: The proposed framework integrates an attention-enhanced deep convolutional encoder–decoder network with a genetic algorithm for seam optimization. The encoder–decoder network generates a structure-aware energy map by jointly modeling spatial, channel, and directional information, enabling a more explicit representation of geometrically coherent regions. Based on this learned representation, the genetic algorithm performs keypoint-constrained seam path optimization, discouraging seam trajectories that intersect structurally critical areas and improving seam continuity. The proposed method is evaluated on three benchmark datasets—UCID, BOSSBase, and BOWS-2—under horizontal, vertical, and mixed seam-removal settings using standard image-quality and saliency-preservation metrics.
Results: Under the horizontal-retargeting baseline-comparison protocol, the proposed method achieves the highest SSIM among the evaluated methods. Compared with the strongest baseline on each dataset, the proposed method improves SSIM by 0.2–1.2 percentage points, PSNR by 0.1–0.9 dB, and saliency preservation by 0.2–0.3 percentage points. A resolution-stratified evaluation from 256×256 to 1024×1024 further shows limited across-resolution variation, with maximum observed ranges of 0.6 percentage points in SSIM, 0.6 dB in PSNR, and 0.8 percentage points in saliency preservation across the evaluated datasets.
Conclusion: Although the incorporation of genetic optimization introduces additional computational cost, the results indicate that combining attention-based energy estimation with optimization-driven seam selection improves structural consistency under the evaluated resizing settings. These findings suggest that the proposed hybrid framework can provide a more reliable balance between saliency preservation, structural continuity, and seam-path optimization in content-aware image retargeting.

Keywords

• Deep Neural Networks
• Convolutional Encoder-Decoder Neural Network
• Seam Carving, Metaheuristic Algorithm

Main Subjects

• Image Processing

Open Access

This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit: http://creativecommons.org/licenses/by/4.0/

 

Publisher’s Note

JECEI Publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

 

Publisher

Shahid Rajaee Teacher Training University