Advanced Digital Image and Signal Processing Techniques for Real-Time Anomaly Detection

Abstract

Real-time anomaly detection is a core requirement in industrial monitoring, surveillance, biomedical diagnostics and predictive maintenance, where unusual events must be identified within strict latency limits. This paper presents a unified framework that combines digital image processing (DIP) and digital signal processing (DSP) techniques for detecting anomalies in streaming visual and one-dimensional sensor data. The framework consists of modality-specific pre-processing, spatial and time–frequency feature extraction, feature-level fusion, unsupervised anomaly scoring and an adaptive decision threshold. Techniques reviewed and integrated include adaptive background modelling, optical flow, edge and texture analysis, the short-time Fourier transform, wavelet decomposition, statistical indicators, autoencoder reconstruction error and isolation-based scoring. Real-time design considerations such as sliding-window processing, computational complexity and embedded deployment are discussed, and an evaluation protocol based on precision, recall, F1-score, ROC-AUC and detection latency is defined. The framework is demonstrated on reproducible synthetic vibration-signal and surface-image data; the signal pipeline achieved an F1-score of 0.949 with a per-window processing time of about 2.0 ms, and the image pipeline achieved a recall of 1.000 at about 1.6 ms per frame on a general-purpose processor. A qualitative comparison of methods and a set of design guidelines are provided to help practitioners select a suitable technique for a given application and hardware budget.

Introduction

The text presents a unified real-time anomaly detection framework that combines Digital Image Processing (DIP) and Digital Signal Processing (DSP) to identify unusual patterns in images, videos, and sensor signals. Traditional anomaly detection methods often analyze visual and signal data separately, limiting their ability to detect complex faults. The proposed framework addresses this limitation through a modular system that processes both data types and combines their features for more accurate and reliable detection.

The literature review covers statistical methods, image-processing techniques, signal analysis, and machine-learning approaches, including background subtraction, optical flow, Fourier and wavelet transforms, Isolation Forests, and autoencoders. The proposed system includes image preprocessing and feature extraction, signal filtering and feature analysis, and feature-level fusion. Image features identify unusual motion, edges, textures, and surface defects, while signal features such as RMS, kurtosis, crest factor, and spectral entropy help detect abnormal operating conditions.

The framework is designed for real-time applications by considering processing speed, memory usage, computational cost, false-alarm rates, and detection delays. Its modular architecture allows different techniques to be selected according to the application and available hardware, while fallback mechanisms support operation when one sensor stream becomes unavailable.

Conclusion

This paper presented a modular framework that combines digital image processing and digital signal processing for real-time anomaly detection. By pairing spatial and time–frequency features, fusing them, and applying unsupervised scoring with an adaptive threshold, the framework can serve varied applications while meeting latency constraints. A demonstration on reproducible synthetic data showed that a Mahalanobis-distance detector on eight signal features reached an F1-score of 0.949 with a processing time of about 2.0 ms per window, and that a background-residual image detector found every defective surface image at about 1.6 ms per frame, albeit with a lower precision of 0.541. The design guidelines, implementation considerations and evaluation protocol provide a practical basis for researchers and engineers. Validation on real data and on embedded hardware will be the focus of subsequent work.

References

[1] V. Chandola, A. Banerjee and V. Kumar, “Anomaly detection: A survey,” ACM Computing Surveys, vol. 41, no. 3, pp. 1–58, 2009. [2] R. C. Gonzalez and R. E. Woods, Digital Image Processing, 4th ed. New York, NY, USA: Pearson, 2018. [3] A. V. Oppenheim and R. W. Schafer, Discrete-Time Signal Processing, 3rd ed. Upper Saddle River, NJ, USA: Pearson, 2010. [4] S. Mallat, A Wavelet Tour of Signal Processing, 3rd ed. Burlington, MA, USA: Academic Press, 2009. [5] C. Stauffer and W. E. L. Grimson, “Adaptive background mixture models for real-time tracking,” in Proc. IEEE CVPR, 1999, pp. 246–252. [6] F. T. Liu, K. M. Ting and Z.-H. Zhou, “Isolation forest,” in Proc. IEEE ICDM, 2008, pp. 413–422. [7] G. Pang, C. Shen, L. Cao and A. van den Hengel, “Deep learning for anomaly detection: A review,” ACM Computing Surveys, vol. 54, no. 2, pp. 1–38, 2021. [8] M. Sakurada and T. Yairi, “Anomaly detection using autoencoders with nonlinear dimensionality reduction,” in Proc. MLSDA Workshop, 2014, pp. 4–11. [9] M. Hasan, J. Choi, J. Neumann, A. K. Roy-Chowdhury and L. S. Davis, “Learning temporal regularity in video sequences,” in Proc. IEEE CVPR, 2016, pp. 733–742. [10] B. Schölkopf, J. C. Platt, J. Shawe-Taylor, A. J. Smola and R. C. Williamson, “Estimating the support of a high-dimensional distribution,” Neural Computation, vol. 13, no. 7, pp. 1443–1471, 2001. [11] I. Daubechies, Ten Lectures on Wavelets. Philadelphia, PA, USA: SIAM, 1992. [12] J. Canny, “A computational approach to edge detection,” IEEE Trans. Pattern Anal. Mach. Intell., vol. PAMI-8, no. 6, pp. 679–698, 1986. [13] B. K. P. Horn and B. G. Schunck, “Determining optical flow,” Artificial Intelligence, vol. 17, nos. 1–3, pp. 185–203, 1981.

Copyright

Copyright © 2026 Monika Kailas Khairnar. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.