METHOD AND APPARATUS FOR FACE RECOGNITION ROBUST TO ALIGNMENT SHAPE OF THE FACE
Provided are a method for performing face recognition and a device for performing face recognition. The method for performing face recognition comprises: a step of estimating prior information of a face shape using a first deep neural network (DNN) of a face recognition device from an input image wh...
Gespeichert in:
Hauptverfasser: | , , , , , , , |
---|---|
Format: | Patent |
Sprache: | eng ; kor |
Schlagworte: | |
Online-Zugang: | Volltext bestellen |
Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
Zusammenfassung: | Provided are a method for performing face recognition and a device for performing face recognition. The method for performing face recognition comprises: a step of estimating prior information of a face shape using a first deep neural network (DNN) of a face recognition device from an input image which is cropped from an image including a face; a step of extracting feature information of the face shape using a second DNN of the face recognition device from the input image; a step of training the face recognition device so that a face image close to an optimally aligned face image can be decoded, based on the prior information and feature information using a face image decoder of the face recognition device; and a step of extracting a face shape-recognized feature from a target image in an inference step using the second DNN included in the face recognition device which is trained. Therefore, the present invention has an effect of providing the method for performing face recognition and the device therefor, capable of extracting face shape information and facial feature information from augmented data using different deep networks, after augmenting data of a face area, and using a face recognition model created by learning feature information obtained by combining the face shape information and the facial feature information to perform face recognition robust to an alignment shape of a face.
얼굴을 포함하는 이미지에서 크롭된 입력 이미지로부터 얼굴 인식 장치의 제1 심층 신경망(Deep Neural Network, DNN)을 사용하여 얼굴 형태의 사전 정보를 추정하는 단계, 입력 이미지로부터 얼굴 인식 장치의 제2 심층 신경망을 사용하여 얼굴 모양의 특징 정보를 추출하는 단계, 얼굴 인식 장치의 얼굴 이미지 디코더를 사용하여 사전 정보 및 특징 정보를 바탕으로, 최적 정렬 얼굴 이미지에 근접한 얼굴 이미지가 디코딩될 수 있도록 얼굴 인식 장치를 트레이닝하는 단계, 그리고 트레이닝된 얼굴 인식 장치에 포함된 제2 심층 신경망을 사용하여 추론 단계에서 목표 이미지로부터 얼굴 형태가 인지된 특징을 추출하는 단계를 통해 얼굴 인식을 수행하는 방법 및 장치가 제공된다. |
---|