Fast image classification algorithms based on sequential analysis

A. V. Savchenko

Zapiski Nauchnykh Seminarov POMI, Investigations on applied mathematics and informatics. Part I, Tome 499 (2021), pp. 267-283

Résumé

In this paper fast image recognition techniques based on statistical sequential analysis are discussed. We examine the possibility to sequentially process the principal components and organize a convolutional neural network with early exits. Particular attention is paid to sequentially learn multi-task lightweight neural network model to predict several facial attributes (age, gender and ethnicity) based on preliminary training on the face classification task. It is highlighted that the whole above-mentioned model should be fine-tuned in order to deal with emotion recognition problem. Experimental study on several datasets demonstrate that the proposed approach is rather accurate and has very low run-time and space complexity when compared to known state-of-the-art methods.