Multimedia Data Recovery and Latent Factor Analysis

Soroor Ghandali

¹Department of Applied Data Science, San Jose State University, San Jose, CA, USA

*Corresponding author

Linear factor models (LFMs) provide a principled framework for discovering low-dimensional latent structures underlying high-dimensional multimedia data. This paper presents a mask-aware Expectation-Maximization (EM) formulation for estimating LFM/factor-analysis parameters when different images, or different image patches, have different missing-pixel locations. The probabilistic latent-variable model itself is classical; the contribution of this work is an explicit image-oriented estimation and recovery workflow in which the observation masks are used in both the E-step and a row-wise M-step. In particular, each pixel’s mean, loading-row parameters, and diagonal noise variance are estimated only from samples where that pixel is actually observed. The paper also clarifies the relationship to EM-FA, EM-PCA, probabilistic PCA, and matrix-completion methods, specifies a low-rank implementation of the observed-data likelihood, and provides a reproducible experimental protocol including initialization, stopping criteria, random-mask generation, classifier specification, repeated trials, and factor-ranking stability checks. Experiments on Camera Man patches and MNIST handwritten digits characterize reconstruction quality and classification accuracy under increasing missing-pixel ratios, with repeated random-mask and initialization trials reported as means and standard deviations. Latent-factor indices are interpreted only after component alignment and stability assessment across random initializations.

Multimedia data recovery, linear factor model, factor analysis, expectation-maximization, missing pixels, latent representation

Soroor Ghandali and Shih Yu Chang (2026). Multimedia Data Recovery and Latent Factor Analysis. Journal of Networking and Network Applications, Volume 6, Issue 2, pp. 67–73. https://doi.org/10.33969/J-NaNA.2026.060203.

[1] Y. Wu, X. Wang, P. Angueira, J. Montalban, N. Hur, S. Ahn, and S.-I. Park, “Local content insertion in the SFN environment for geo-graphically segmented localcasting,” in Proceedings of IEEE Interna-tional Symposium on Broadband Multimedia Systems and Broadcasting (BMSB), Jun. 2025, pp. 1–6.

[2] S. Ahn, B.-m. Lim, J. Kim, C. E. C. Ribeiro, L. Fausto, and S.-I. Park, “Geographic segmented localcasting based on LDM: Design and plans in Brazil,” in Proceedings of IEEE International Symposium on Broadband Multimedia Systems and Broadcasting (BMSB), Jun. 2025, pp. 1–4.

[3] L. N. D. S. Junior, R. S. Rabaca, G. H. M. G. de Oliveira, G. L. Montalvao, G. A. Chiesa, C. Akamine, B.-M. Lim, S.-I. Park, and

Y. Wu, “TV 3.0 geographical segmentation by over-the-air physical layer,” in Proceedings of IEEE International Symposium on Broadband Multimedia Systems and Broadcasting (BMSB), Jun. 2025, pp. 1–4.

[4] G. C. Livanos and V. Anishchenko, “Development of an ultra-long-range wireless backhaul solution using ATSC 3.0,” IEEE Transactions on Broadcasting, vol. 69, no. 2, pp. 629–634, May 2023.

[5] Q. Gao, Z. Li, J. Li, J. Xue, and Y. Xing, “Mobile computing force network architecture and quality of service (QoS) mechanism to-wards network-computing integration,” in Proceedings of IEEE Interna-tional Symposium on Broadband Multimedia Systems and Broadcasting (BMSB), Jun. 2025, pp. 1–6.

[6] D. J. Bartholomew, M. Knott, and I. Moustaki, Latent Variable Models and Factor Analysis: A Unified Approach. John Wiley & Sons, 2011.

[7] M. E. Tipping and C. M. Bishop, “Probabilistic principal component analysis,” Journal of the Royal Statistical Society: Series B (Statistical Methodology), vol. 61, no. 3, pp. 611–622, Sep. 1999.

[8] Z. Ghahramani and G. E. Hinton, “The EM algorithm for factor analysis,” University of Toronto Technical Report CRG-TR-96-1, 1996.

[9] S. T. Roweis, “EM algorithms for PCA and SPCA,” in Advances in Neural Information Processing Systems, vol. 10, 1998, pp. 626–632.

[10] E. J. Candes and B. Recht, “Exact matrix completion via convex optimization,” Foundations of Computational Mathematics, vol. 9, no. 6, pp. 717–772, 2009.

[11] R. Mazumder, T. Hastie, and R. Tibshirani, “Spectral regularization algorithms for learning large incomplete matrices,” Journal of Machine Learning Research, vol. 11, pp. 2287–2322, 2010.

[12] C. M. Bishop, “Latent variable models,” in Learning in Graphical Models. Springer, 1998, pp. 371–403.

[13] C. M. Bishop, Pattern Recognition and Machine Learning. New York, NY: Springer, 2006.

[14] S. van der Walt et al., “scikit-image: Image processing in Python,” 2014.

[15] Y. LeCun, C. Cortes, and C. Burges, “The MNIST database of handwrit-ten digits,” http://yann.lecun.com/exdb/mnist/, 1998, accessed: 2025-10-31.