利用多工學習加強人臉辨識與特徵偵測

Abstract

近年來，卷積神經網路在人臉表示法的學習中有非常傑出的表現與成果，但大部分的研究專注於利用大量的資料學習人臉表示法而非同時利用人臉最具有語意的特徵如性別、年齡與膚色等來更佳化人臉表示法。在這篇論文中，我們提出使用多工學習的卷積神經網路來加強人臉辨識與特徵偵測。更精確的說，我們專注於同時學習人臉表示法與特徵偵測，並同時解決這兩種問題。在實驗中，使用大量的人臉圖片伴隨身分、性別、年齡等標籤，用設計出的多工卷積神經網路架構學習模型，接著使用LFW以即Adience兩個數據集來評估所學的結果，並且和傳統的局部二值模式(LBP)以即單一學習模型作比較，發現多工學習對於人臉辨識和特徵偵測有所幫助。

Convolution neural network (CNN) has been shown as the state-of-the-art approach for learning face representations in recent years. However, previous works only utilized identity information instead of leveraging human attributes (e.g., gender and age) which contain high-level semantic meanings to learn robuster features. In this work, we aim to learn discriminative features to improve face recognition through multi-task learning with human attributes. Specifically, we focus on simultaneously optimizing face recognition and human attributes estimation. In our experiments, we learn face representation by training the largest publicly face dataset CASIA-WebFace with gender and age label, and then evaluate learned features on widely-used LFW benchmark for face verification and identification. We also compare the effectiveness of different attributes for identification. The results show that the proposed model outperforms hand-crafted feature such as high-dimensional LBP, and human attributes really provide useful semantic cues. We also do experiments on gender and age estimation on Adience benchmark to justify that human attribute prediction can also benefit from rich identity information.