• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.14 no.10
• /
• pp.4080-4097
• /
• 2020

In recent years, convolutional neural network (CNN) has become the primary method for face detection. But its shortcomings are obvious, such as expensive calculation, heavy model, etc. This makes CNN difficult to use on the mobile devices which have limited computing and storage capabilities. Therefore, the design of lightweight CNN for face detection is becoming more and more important with the popularity of smartphones and mobile Internet. Based on the CPU real-time face detector FaceBoxes, we propose a multi-task lightweight face detector, which has low computing cost and higher detection precision. First, to improve the detection capability, the squeeze and excitation modules are used to extract attention between channels. Then, the textual and semantic information are extracted by shallow networks and deep networks respectively to get rich features. Finally, the landmark detection module is used to improve the detection performance for small faces and provide landmark data for face alignment. Experiments on AFW, FDDB, PASCAL, and WIDER FACE datasets show that our algorithm has achieved significant improvement in the mean average precision. Especially, on the WIDER FACE hard validation set, our algorithm outperforms the mean average precision of FaceBoxes by 7.2%. For VGA-resolution images, the running speed of our algorithm can reach 23FPS on a CPU device.

• Journal of Korea Multimedia Society
• /
• v.24 no.8
• /
• pp.1020-1025
• /
• 2021

Facial alignment is very important task for human life. And facial landmark detection is one of the instrumental methods in face alignment. We introduce the stacked hourglass networks with transposed convolutional layers for facial landmark detection. our method substitutes nearest neighbor upsampling for transposed convolutional layer. Our method returns better accuracy in facial landmark detection compared to stacked hourglass networks with nearest neighbor upsampling.

• Journal of Broadcast Engineering
• /
• v.24 no.4
• /
• pp.564-572
• /
• 2019

With the development of deep learning, the performance of facial landmark detection methods has been greatly improved. The heat map regression method, which is a representative facial landmark detection method, is widely used as an efficient and robust method. However, the landmark coordinates cannot be directly obtained through a single network, and the accuracy is reduced in determining the landmark coordinates from the heat map. To solve these problems, we propose to combine integral regression with the existing heat map regression method. Through experiments using various datasets, we show that the proposed integral regression network significantly improves the performance of facial landmark detection.

• Journal of the Institute of Electronics and Information Engineers
• /
• v.50 no.5
• /
• pp.199-207
• /
• 2013

Spin images representing efficiently surface features of 3D mesh models have been used to detect facial landmark points. However, at a certain point, different normal direction can lead to quite different spin images. Moreover, since 3D points are projected to the 2D (${\alpha}-{\beta}$) space during spin image generation, surface features cannot be described clearly. In this paper, we present a method to detect 3D facial landmark using improved spin images by partitioning the search area with respect to angle. By generating sub-spin images for angular partitioned 3D spaces, more unique features describing corresponding surfaces can be obtained, and improve the performance of landmark detection. In order to generate spin images robust to inaccurate surface normal direction, we utilize on averaging surface normal with its neighboring normal vectors. The experimental results show that the proposed method increases the accuracy in landmark detection by about 34% over a conventional method.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.22 no.1
• /
• pp.163-170
• /
• 2022

This paper introduces a method of extracting facial features due to unrefined inputs in real life and improving the problem of not guaranteeing the ideal performance and speed of the object recognition model through a storage algorithm through weight summation. Many facial recognition processes ensure accuracy in ideal situations, but the problem of not being able to cope with numerous biases that can occur in real life is drawing attention, which may soon lead to serious problems in the face recognition process closely related to security. This paper presents a method of quickly and accurately recognizing faces in real time by comparing feature points extracted as input with a small number of feature points that are not overfit to multiple biases, using that various variables such as picture composition eventually take an average form.

• Journal of Advanced Navigation Technology
• /
• v.24 no.5
• /
• pp.430-437
• /
• 2020

The proposed model implements a model that improves the face prediction rate and recognition rate through learning with an artificial neural network using face detection, landmark and face recognition algorithms. After landmarking in the face images of a specific person, the proposed model use the previously learned Caffe model to extract face detection and embedding vector 128D. The learning is learned by building machine learning algorithms such as support vector machine (SVM) and deep neural network (DNN). Face recognition is tested with a face image different from the learned figure using the learned model. As a result of the experiment, the result of learning with DNN rather than SVM showed better prediction rate and recognition rate. However, when the hidden layer of DNN is increased, the prediction rate increases but the recognition rate decreases. This is judged as overfitting caused by a small number of objects to be recognized. As a result of learning by adding a clear face image to the proposed model, it is confirmed that the result of high prediction rate and recognition rate can be obtained. This research will be able to obtain better recognition and prediction rates through effective deep learning establishment by utilizing more face image data.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 2007.04a
• /
• pp.143-146
• /
• 2007

객체의 특징점을 추출할 때, 일반적으로 모델 기반 접근을 사용한다. 본 논문에서는 이러한 모델 기반 특징점 추출 알고리즘으로 PCA를 근간으로 하는 Active Appearance Model을 이용하는데, 기존의 AAM 알고리즘은 모든 특징점을 하나의 군집으로 기준하여 PCA를 수행하지만 본 논문에서는 이것을 각 주요 부위별 학습 모델로 분리하여 수행한다. 그리고 이러한 모델에서 특징점을 찾을 때, 발산하는 문제에 빠지지 않기 위한 방법을 제시한다. 제시한 방법의 모델을 이용하여 실험 할 경우의 결과와 이를 통한 개별 모델의 특성에 대하여 파악한 결과를 제시한다.

• Journal of Multimedia Information System
• /
• v.5 no.1
• /
• pp.15-20
• /
• 2018

There are a lot of problems in the face detection area. One of them is detecting faces by facial features and reducing number of the false negatives and positions. This paper is directed to solve this problem by the proposed triangle method. Also, this paper explans cascades, Haar-like features, AdaBoost, HOG. We propose a scheme using 12-net, 24-net, 48-net to scan images and improve efficiency. Using triangle method for frontal pose, B and B1 methods for other poses in neural networks are proposed.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.21 no.9
• /
• pp.25-31
• /
• 2020

In this study, for the safe use of a portable ultrasonic skin-beauty device, an android app was developed to show the irradiation locations of focused ultrasound to a user through augmented reality (AR) and enable stable self-surgery. The utility of the app was assessed through testing. While the user is making a facial treatment with the beauty device, the user's face and the ultrasonic irradiation location on the face are detected in real-time with a smart-phone camera. The irradiation location is then indicated on the face image and shown to the user so that excessive ultrasound is not irradiated to the same area during treatment. To this end, ML-Kit is used to detect the user's face landmarks in real-time, and they are compared with a reference face model to estimate the pose of the face, such as rotation and movement. After mounting a LED on the ultrasonic irradiation part of the device and operating the LED during irradiation, the LED light was searched to find the position of the ultrasonic irradiation on the smart-phone screen, and the irradiation position was registered and displayed on the face image based on the estimated face pose. Each task performed in the app was implemented through the thread and the timer, and all tasks were executed within 75 ms. The test results showed that the time taken to register and display 120 ultrasound irradiation positions was less than 25ms, and the display accuracy was within 20mm when the face did not rotate significantly.

• Journal of Korea Multimedia Society
• /
• v.20 no.11
• /
• pp.1720-1728
• /
• 2017

Active shape model is widely used in the field of image processing especially on arbitrary meaningful shape extraction from single gray level image. Cootes et. al. showed efficient detection of variable shape from image by using covariance and mean shape from learning. There are two stages of learning and testing. Hahn applied enhanced shape alignment method rather than using Cootes's rotation and scale scheme. Hahn did not modified the profile itself. In this paper, the method using directional one dimensional profile is proposed to enhance Cootes's one dimensional profile and the shape alignment algorithm of Hahn is combined. The performance of the proposed method was superior to Cootes's and Hahn's. Average landmark estimation error for each image was 27.72 pixels and 39.46 for Cootes's and 33.73 for Hahn's each.