Version 4.1

Presentation: AI Face Recognition with OpenCV DNN module and application in digiKam photo management software


Currently, we are observing an incredible development in technologies, especially in Artificial Intelligence field. Indeed, by learning from massive data, AI is particularly good at some tasks that normal algorithms cannot achieve as good level of performance, such as: image classification, speech recognition, object detection, tendency prediction, feature extraction, etc. Moreover, new AI algorithms with the emergence of neural networks and deep learning even makes AI models more robust, so that they can now give better prediction without any limitation in improving themselves.

Being aware of those assets, digiKam team has considered using deep learning in digiKam. Thus, this presentation aims to introduce a new implementation of facial recognition in digiKam, based on deep learning models and OpenCV DNN module, so as to improve the performance of facial recognition module.

For users, facial recognition is one of the most interesting features in digiKam. However, it was not robust enough and computationally expensive, so users usually did not totally trust and handle digiKam to automatically recognize faces in their photos. Inheriting from previous work on facial recognition in digiKam, new approach implementing neural-network-based solution with OpenCV DNN module, exploiting forward operation on the network, is expected to deliver better computational performance, while keeping the outstanding accuracy of face engine.

Hence, this presentation is intended to discuss in details current neural network models being studied for facial recognition, the advantages and disadvantages of most popular models. In addition, implementation details with OpenCV DNN module, as well as evaluation metrics and final benchmark scores will be also presented at the end to evaluate the work and the performance of that approach.