基於簡單資料擴增學習之快速非共平面人臉偵測

Abstract

隨著行動裝置的普及化，各式相關應用不斷地被開發以滿足使用者所需。然而受限於可攜性要求之先天限制，無法無止盡地擴增其計算能力去應付眾多需求。基於此一考量，本論文針對人臉偵測提出一種新的訓練方式，藉由簡單資料擴增的學習模式讓行動裝置在僅增加少額計算量的情形下強化非共平面人臉的偵測能力，經由多種數據集的實驗比對確認為可行做法。其有效率的功能強化在提升性能的同時亦保留計算能力允許行動裝置同時執行更多功能以利開發更多豐富的應用。

With the popularity of mobile devices, all kinds of related applications are constantly being developed to meet user needs. However, subject to the requirements of portability restrictions, we can not endlessly amplified its computing performance to cope with the many demands. Based on this consideration, this paper proposes a new training methods for face detection. The learning by simple data augmentation allows mobile devices to strengthen the out-of-plane face detection capabilities with little increase of computed consumption. According to the experiments of various data sets, the proposed method is recognized as a viable approach. The functional enhancement with efficiency not only improves the performance but also allows mobile devices to perform more functions simultaneously to facilitate the development of more rich applications.