Bridging Video Compression and Action Recognition via Task-Aware Progressive SPIHT

Dr. Vipparthy Bhagya Raju ^1,2, Sarath Chandra Veerla ^1*, Kasa Ravindra ³

¹School of Sciences and Humanities, SR University, Warangal, 506371, Telangana, India.

²Department of Electronics and Communication Engineering, Siddhartha Institute of Engineering and Technology, Ibrahimpatnam, Hyderabad, 501506, Telangana, India.

^1*School of Sciences and Humanities, SR University, Warangal, 506371, Telangana, India.
Email: sarathchandra.veerla85@gmail.com
ORCID: https://orcid.org/0000-0001-9288-9107

³Professor and Director, St Martin's Engineering College, Secunderabad, Telangana, India.
Email: drkasaravindra@gmail.com

Corresponding Author: Sarath Chandra Veerla, School of Sciences and Humanities, SR University, Warangal, 506371, Telangana, India. Email: sarathchandra.veerla85@gmail.com

V. Bhagya Raju ORCID: https://orcid.org/0000-0001-6781-0639

---

ABSTRACT

Human Action Recognition (HAR) systems deployed in real-time surveillance, edge computing, and bandwidth-constrained environments require efficient video compression without sacrificing recognition accuracy. Conventional compression schemes such as Set Partitioning in Hierarchical Trees (SPIHT) are optimized for pixel-level fidelity metrics like PSNR and SSIM, which do not necessarily preserve motion dynamics, spatio-temporal edges, or skeletal structures critical for action recognition. This paper proposes a Task-Aware Progressive SPIHT (TA-PSPIHT) framework that bridges video compression and action recognition by aligning encoding priorities with task relevance rather than visual reconstruction quality alone.

The proposed method integrates lightweight pose estimation and optical-flow magnitude maps to generate an importance mask that identifies motion- and skeleton-dominant regions. This mask is incorporated into the SPIHT set-partitioning mechanism through Weighted Significance Testing, enabling action-relevant wavelet coefficients to be encoded earlier in the progressive bitstream. Furthermore, a 3D Temporal-Priority SPIHT structure exploits spatio-temporal dependencies across frames, while a Policy-Gradient–based Bit-Dropping strategy optimizes rate–recognition trade-offs.

Experimental results demonstrate that the proposed framework significantly improves action recognition accuracy at low bitrates compared to conventional SPIHT, while maintaining computational efficiency and progressive transmission capability. The proposed approach provides a practical and scalable solution for task-driven video analytics in resource-constrained environments.

Keywords: Task-aware video compression, Progressive SPIHT, Human Action Recognition (HAR), Motion-guided encoding, Pose-aware compression, Optical flow, 3D wavelet coding, Rate–recognition optimization, Edge video analytics, Progressive transmission.

How to cite this article: Raju VB, Veerla SC, Ravindra K. Bridging video compression and action recognition via task-aware progressive SPIHT. Int J Drug Deliv Technol. 2026;16(3s): 860-866; DOI: 10.25258/ijddt.16.3s.104

Source of support: Nil.

Conflict of interest: None