The problem of automatically detecting and recognizing faces has received significant attention in the past fifty years and systems based on Convolutional Neural Networks (CNNs) or convolutional architectures currently represent the state of the art. In this paper, the authors tackle the problem of optimizing the energy-rate-accuracy characteristics of a convolutional architecture for face recognition in distributed battery-operated devices, with the objective of enabling the Analyze-Then-Compress paradigm for face identification/recognition.
The proposed architecture requires on average 94% less energy as compared to the baseline architecture in order to extract features from an input image on a low-power ARM-based Raspberry Pi computer. Also, it achieves a recognition rate of 85.41% (only 1.72% less than the uncompressed baseline) while requiring as few as 7379 bits per image.
L. Bondi, L. Baroffio, M. Cesana, M. Tagliasacchi, G. Chiachia, A. Rocha. Rate-energy-accuracy optimization of convolutional architectures for face recognition. Journal of Visual Communication and Image Representation, v. 36, 2016, 142-148. http://dx.doi.org/10.1016/j.jvcir.2015.12.015.