Design of Face Detection Based on Optimized MTCNN :최적화 된 MTCNN에 기반한 얼굴 인식 디지인

Yumei Fu

추천

검색

자료유형: 학위논문

저자정보: Yumei Fu (동의대학교, 동의대학교 대학원)

지도교수: 장종욱

발행연도: 2020

저작권: 동의대학교 논문은 저작권에 의해 보호받습니다.

이용수4

이 논문의 연구 히스토리 (3)

2020

Research on Improved MTCNN Model Algorithm for Face Detection

Yumei Fu , Minyoung Kim , Jongwook Jang 한국정보통신학회 종합학술대회 논문집 2020.07 학술대회자료

Design of Face Detection Based on Optimized MTCNN

Yumei Fu 컴퓨터공학과 2020.01 학위논문

2019

A Study on Face Detection Algorithm Implementation Based on MTCNN Model for Complex Environments

Yumei Fu , Minyoung Kim , Jong-wook Jang 한국정보통신학회 종합학술대회 논문집 2019.10 학술대회자료

이 논문의 후속연구가 궁금하신가요?
연관 학술논문 또는 학술발표를 통해 보다 발전된 연구결과를 확인하실 수 있습니다.
이 논문의 연구 히스토리 확인하기

초록· 키워드

오류제보하기

오늘날 빅 데이터 시대가 도래하면서 사물 인터넷의 급속한 발전과 지능형 단말기의 광범위한 인기로 머신 비전 기술의 급속한 발전이 크게 촉진되었다. 머신 비전의 중요한 부분으로, 얼굴 인식은 전 세계의 학자들에게 널리 주목을 받았다. 현재, 얼굴인식에 대한 연구는 주로 비디오 이미지에 기초한 얼굴 인식 및 정태적인 이미지에 기초한 얼굴 인식을 포함한다. 이 논문은 주로 정태(靜態)적인 이미지에 기반한 얼굴 인식 알고리즘에 대해 설명한다. 정태적인 이미지에서 사람 얼굴의 존재를 감지하고 사람 얼굴을 정확하게 찾고 이미지에서 사람 얼굴 수를 계산할 수 있다. 정태적인 이미지를 기반으로 한 얼굴 감지의 어려움은 빛, 각도 및 얼굴 표정의 변화로 인해 이미지의 얼굴이 바뀌어 흐려지고 얼굴 테두리가 배경과 융합(融合)되는 것이다. 기존의 검출 방법에서 주요 특징 추출 방법은 주요 성분 분석, Gabor 변환, HOG 방향 기울기 히스토그램 특징 등이 있다. 이러한 특징 추출 알고리즘은 간단한 배경 환경에서 좋은 결과를 얻을 수 있다. 그러나 빅데이터 시대에는 감지해야 할 얼굴의 수가 많고, 각 얼굴의 배경 환경이 복잡하여 얼굴 특징을 정확하게 추출할 수 없어 기존의 얼굴 인식의 프로세스는 여기에 사용하기에 느리다.
DCNN(Deep Convolutional Neural Network)에는 비선형적으로 설명 기능이 있다. 다차원의 복잡한 데이터 구조에서 효과적인 기능을 추출하고 얼굴감지의 정확도와 속도를 크게 향상 시키며 인터넷에서 얼굴 감지 기술의 적용을 가속화 할 수 있다.
본 논문은 TensorFlow 신경망을 기반으로 개선 된 MTCNN(Multi-task Convolutional Neural Network) 얼굴 감지 알고리즘을 사용하고 DCNN의 구조를 이용하며 MTCNN 모델의 네트워크 깊이를 깊게하고 컨볼 루션 커널 및 풀을 수정할 수 있다. 커널의 크기를 줄이고 보다 효과적인 특징 값으로, 최적화된 MTCNN 모델은 얼굴 특징 추출의 정확도를 향상시키는 것으로 입증되었으며, 이는 이전의 MTCNN 정확도보다는 0.8 %로 높게 측정되고 최종적으로 98.8 %에 이르렀다.

Nowadays, with the advent of the era of big data, the rapid development of the Internet of Things and the widespread popularity of intelligent terminals have greatly promoted the rapid development of machine vision technology. As an important part of machine vision, face detection has been widely concerned by scholars all over the world. At present, the research of face detection mainly includes face detection based on video image and face detection based on static image. This paper mainly discusses face detection algorithm based on static image. It can detect the existence of human face in the static image, locate the human face accurately and calculate the number of human faces in the image. The difficulty of object detection based on static pictures is that the faces in the picture will change due to changes in light, angle, and facial expressions, which will cause the faces in the image to be blurred, the face borders to merge with the background or the face Obscured by objects, etc. In the traditional detection methods, the main feature extraction methods are principal component analysis, Gabor transform, HOG directional gradient histogram features and so on. These feature extraction algorithms have good results in simple scenes. However, in the era of big data, the number of faces to be detected is huge and the background environment of each face is complex, which leads to the inability to extract face features accurately, the ability of generalization is reduced, and the face detection process is slow.
DCNN has the ability of non-linear description. It can extract effective features in multi-dimensional complex data structure, greatly improve the accuracy and speed of face detection, and accelerate the application of
face detection technology on the Internet of things.
This paper is based on the TensorFlow neural network, using an improved MTCNN face detection algorithm, drawing on the structure of the DCNN, deepening the network depth of the MTCNN model, and modifying the convolution kernel and pool Reduce the size of the kernel, obtain more effective feature values, It is proved that the optimized MTCNN model improves the accuracy of facial feature extraction, which is 0.8% higher than the previous MTCNN accuracy, reaching 98.8%.

1. Introduction ·············································································································1
1.1 Purpose ···············································································································1
1.2 Motivation ·········································································································2
1.3 Methodology of the study ············································································2
1.4 Thesis Outline ··································································································3
2. Background ·············································································································5
2.1 Hardware Configuration ················································································5
2.2 TensorFlow Framework ················································································6
2.3 Basic Knowledge of CNN ············································································8
2.3.1 Convolutional Layer ·········································································10
2.3.2 Pooling Layer ··························································································12
2.3.3 Fully Connected Layer ·········································································13
2.3.4 Related Parameters and Functions in CNN ··································· 14
2.3.5 Training Process ····················································································16
2.4 Pervious MTCNN ·························································································16
2.4.1 Image Pyramid ························································································18
2.4.2 Model Structure ······················································································19
2.4.3 Key Technology ···················································································21
2.4.3.1 Face ClassificatSion Algorithm ·······················································21
2.4.3.2 Bounding Boxes Regression of Selection ·····································22
2.3.4.4 Facial Landmark Localization Algorithm ······································24
2.3.4.5 Regularization ·······················································································24
3. Optimized Design of MTCNN Model ··························································26
3.1 Optimized MTCNN Face Detection Process ······································· 26
3.2 Optimized CNN Model Structure ······························································29
4. Implementation and Results ··············································································34
4.1 Training Data Sets ·······················································································34
4.2 Image Preprocessing ·····················································································36
4.2.1 Data Enhancement Setting ··································································37
4.2.2 Image Pyramid Setting ········································································38
4.3 Model and parameter settings before training ································· 39
4.5 Test Results and Analysis ·······································································44
4.5.1 Comparative analysis of accuracy and speed ································44
4.5.2 Accuracy in Complex Environment ··················································45
5. Conclusion ·············································································································47
References ··················································································································49

최근 본 자료

전체보기

구분	그룹	데이터 항목
AI 학습용 데이터	원문	원문 PDF 파일
AI 학습용 데이터	원문 + 메타 (기본/상세)	원문 PDF 파일 및 서지정보 CSV
대량 구매용 데이터	B2B 구독 방식	특정 자료 한정으로 원문 접근 권한 부여
대량 구매용 데이터	URL 전달 방식	바로 PDF 뷰어를 열람할 수 있는 URL 제공

구분	그룹	데이터 항목
AI 학습용 데이터	기본 메타	발행기관명, 간행물명, 권호명, 권(vol), 호(issue), 통권, 발행연도, 발행월, 논문명, 저자명, 시작페이지, 종료페이지, 전체페이지, 상세페이지URL
상세 메타 데이터	발행기관 메타	발행기관 이명, 영문명, 창립연도, 홈페이지URL, 발행기관 소개
	간행물 메타	부제목, 간행물 유형, ISSN, ISBN, 최초발행연도, 폐간연도, 간행빈도, 발행주기, 등재사항, 이용수, 피인용수, 권호수, 논문수, 표지이미지
	논문 메타	작성 언어, 부제목, 대등제목, 목차, 키워드, 초록, 이미지, 참고문헌, 이용수, 피인용수, 논문활용도, DBpia통합주제분류, KDC분류, DDC분류, 한국연구재단분류, UCI, DOI
	저자 메타	소속기관, 소속부서, 직급, 연구분야, 연구키워드, 이용수, 피인용수, 저자 논문활용도

구분	그룹	데이터 항목
※ 결합형/맞춤형 메타 데이터는 신청 내용에 따라 다양하게 제공 가능
이용순위 정보	주제분야별 많이 이용된 논문	“인문학”에서 많이 이용된 논문 TOP100
	이용기관별 많이 이용된 논문	“중고등학교”에서 많이 이용된 논문 TOP100
	세부기관별 많이 이용된 논문	“서울대학교”에서 많이 이용된 논문 TOP100
	키워드별 많이 이용된 논문	“Chat GPT”에서 많이 이용된 논문 TOP100
키워드 정보	많이 이용된 키워드	특정기간/분야/저널 내 많이 이용된 키워드
	많이 발행된 키워드	특정기간/분야/저널 내 많이 발행된 키워드
	많이 검색된 키워드	특정기간/분야/저널 내 많이 검색된 키워드
	연구 트렌드 키워드	특정 키워드 연관 연구동향 분석 데이터 키워드

논문 기본 정보

이 논문의 연구 히스토리 (3)

초록· 키워드

목차

최근 본 자료

댓글(0)