Efficient and Scalable Deep Learning Based Face and Object Recognition System




Siddaiah, Vittal

Journal Title

Journal ISSN

Volume Title



Artificial Intelligence (AI) is the panacea for both prescriptive and predictive analytics through Machine Learning (ML) techniques, demands for computational performance, and snowballing over the decades. Pattern Recognition is increasingly demanding in AI applications that include neural networks-based machine learning. In this research, we are dealing face recognition domain of pattern recognition, popularly termed computer vision. Computer vision enables a wide range of applications spanning across industrial, retail, health care, smart cities in robotics/drones, self-driving cars, augmented reality, optical character recognition, face and gesture recognition, smart Internet of Things, portable/wearable electronics, Law enforcement, and much more. Conventional methods like HAAR and HOG algorithms evolved with improved accuracy; these conventional methods were confined and domain-specific and achieved an accuracy of up to 80% in detection. HAAR and HOG-based algorithms demand expert handcrafting in the design to improve accuracy; they are static and non-scalable. In Deep neural networks (DNN), the algorithms are generic and dynamic. DNN learning enables the model to learn from the data. Traditional learning models are saturated regarding the accuracy, while dynamic Learning improves continually over the quantum of training samples. Today there are DNNs in domains that have achieved over 99% accuracy, which is beyond the ground reality. DNN has established itself as a triumphant set of models for learning relevant connotative representations of data. Training of deep-learning models is compute-intensive, and there is an industry-wide trend towards hardware specialization to improve performance. This research uses a DNN-based generic, efficient, scalable, and platform-independent framework that can be extendable across platforms. The proposed framework involves computer vision techniques suitable for unsupervised Learning with low latency and high performance. The proposed framework would be open-source, tested across diverse datasets, compatible and scalable across platforms, with low latency and a small footprint. The framework would serve as a benchmark and publish the rating parameters of response times, latencies, and accuracy that grade and differentiates various platforms.



artificial intelligence, deep learning, object recognition, face recognition


Siddaiah, V. (2023). Efficient and scalable deep learning based face and object recognition system (Unpublished thesis). Texas State University, San Marcos, Texas.


Rights Holder

Rights License

Rights URI