EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

SOP: Selective Orthogonal Projection for Composed Image Retrieval

Su Cheng, Guoyang Liu iD

Sensors · 04.03.2026 · 26 (5) · 1621

DOI: 10.3390/s26051621 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Sensors, online erschienen am 04.03.2026. Verfasst von Su Cheng und 1 weiteren Autor:innen. Beteiligte Einrichtung: School of Integrated Circuits, Shandong University, Jinan 250101, China. Die Arbeit nennt 52 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Natural Science Foundation of China, Natural Science Foundation of Shandong Province, the Guangdong Basic and Applied Basic Research Foundation.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The proliferation of intelligent sensor networks in urban surveillance and remote sensing has triggered the explosive growth of unstructured visual sensor data. Accurately retrieving targets from these massive streams based on complex cross-modal user intents remains a critical bottleneck for efficient intelligent perception. Composed Image Retrieval (CIR) addresses this by enabling retrieval via a multi-modal query that combines a reference image with semantic control signals. However, existing methods often struggle with abstract instructions in real-world scenarios. Consequently, models often suffer from feature distribution shifts due to focus ambiguity, as well as semantic erosion caused by highly entangled visual and textual features. To address these challenges, we propose a geometry-based Selective Orthogonal Projection Network (SOP). First, the Selective Focus Recovery module quantifies instruction uncertainty via information entropy and calibrates shifted query features to the true target distribution using structural consistency regularization. Second, to ensure data fidelity, we introduce Orthogonal Subspace Projectionand Geometric Composition Fidelity. These mechanisms employ Gram–Schmidt orthogonalization to decouple features into a constant visual base and an orthogonal modification increment, restricting semantic modifications to the null space. Extensive experiments on FashionIQ, Shoes, and CIRR datasets demonstrate that SOP significantly outperforms SOTA methods, offering a novel solution for efficient large-scale sensor data retrieval and analysis.

Schlagwörter

Stichwörter der Autor:innen

composed image retrievalcross-modal information retrievalsemantic alignment

Autor:innen und Institutionen

  • School of Integrated Circuits, Shandong University, Jinan 250101, China.
  • Shenzhen Research Institute of Shandong University, Shenzhen 518000, China.

ORCID-Profile: Guoyang Liu

Förderung

  • National Natural Science Foundation of China
  • Natural Science Foundation of Shandong Province
  • the Guangdong Basic and Applied Basic Research Foundation
  • the Shenzhen Fundamental Research Program
  • the Sichuan Provincial Key Laboratory of Philosophy and Social Science for Language Intelligence in Special Education
  • the National Key Research and Development Program of China
  • and the Key Laboratory of Social Computing and Cognitive Intelligence (Dalian University of Technology), Ministry of Education
  • Guangdong Basic and Applied Basic Research Foundation

Zitieren

APA

Cheng, S., & Liu, G. (2026). SOP: Selective Orthogonal Projection for Composed Image Retrieval. Sensors, 26(5), 1621. https://doi.org/10.3390/s26051621

Vancouver

Cheng S, Liu G. SOP: Selective Orthogonal Projection for Composed Image Retrieval. Sensors. 2026;26(5):1621. doi:10.3390/s26051621

BibTeX
@article{Cheng2026,
  author = {Cheng, Su and Liu, Guoyang},
  title = {SOP: Selective Orthogonal Projection for Composed Image Retrieval},
  journal = {Sensors},
  year = {2026},
  volume = {26},
  number = {5},
  pages = {1621},
  doi = {10.3390/s26051621}
}
Kostenlos starten →