• Proceedings of the Korea Information Processing Society Conference
• /
• 2019.10a
• /
• pp.1101-1104
• /
• 2019

키넥트는 RGBD 카메라로 인체의 뼈대와 관절을 3D 공간에서 스켈레톤 데이터수집을 가능하게 해주었다. 스켈레톤 데이터를 활용한 행동 분류는 RNN, CNN 등 다양한 인공 신경망으로 접근하고 있다. 본 연구는 키넥트를 이용해서 Skeleton Joints를 수집하고, DNN 기반 스켈레톤 모델링 학습으로 행동을 분류한다. Skeleton Joints Processing 과정은 키넥트의 Depth Map 기반의 Skeleton Tracker로 25가지 Skeleton Joints 좌표를 얻고, 학습을 위한 전처리 과정으로 각 좌표를 상대좌표로 변경하고 데이터 수를 제한하며, Joint가 트래킹 되지 않은 부분에 대한 예외 처리를 수행한다. 스켈레톤 모델링 학습 과정에선 3계층의 DNN 신경망을 구축하고, softmax_cross_entropy 함수로 Skeleton Joints를 집는 모션, 내려놓는 모션, 팔짱 낀 모션, 얼굴을 가까이 가져가는 모션 해서 4가지 행동으로 분류한다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2019.10a
• /
• pp.1030-1032
• /
• 2019

형태소 분석을 통하여 한국어 문장을 형태소 단위의 임베딩 및 학습 관련 연구가 되었으나 최근 비정형적인 텍스트 데이터의 증가에 따라 음소 단위의 임베딩을 통한 신경망 학습에 대한 요구가 높아지고 있다. 본 논문은 비정형적인 텍스트 감성 분석 성능 향상을 위해 음소 단위의 토큰을 생성하고 이를 CNN 모형을 기반으로 다차원 임베딩을 수행하고 감성분석을 위하여 양방향 순환신경망 모델을 사용하여 유튜브의 비정형 텍스트를 학습시켰다. 그 결과 텍스트의 긍정 부정 판별에 있어 90%의 정확도를 보였다.

• Journal of Information Processing Systems
• /
• v.16 no.4
• /
• pp.820-831
• /
• 2020

Extractive document summarization aims to select a few sentences while preserving its main information on a given document, but the current extractive methods do not consider the sentence-information repeat problem especially for news document summarization. In view of the importance and redundancy of news text information, in this paper, we propose a neural extractive summarization approach with joint sentence semantic clipping and selection, which can effectively solve the problem of news text summary sentence repetition. Specifically, a hierarchical selective encoding network is constructed for both sentence-level and document-level document representations, and data containing important information is extracted on news text; a sentence extractor strategy is then adopted for joint scoring and redundant information clipping. This way, our model strikes a balance between important information extraction and redundant information filtering. Experimental results on both CNN/Daily Mail dataset and Court Public Opinion News dataset we built are presented to show the effectiveness of our proposed approach in terms of ROUGE metrics, especially for redundant information filtering.

• Journal of Information Processing Systems
• /
• v.17 no.2
• /
• pp.385-398
• /
• 2021

The complexity of deep learning models affects the real-time performance of gesture recognition, thereby limiting the application of gesture recognition algorithms in actual scenarios. Hence, a residual learning neural network based on a deep convolutional neural network is proposed. First, small convolution kernels are used to extract the local details of gesture images. Subsequently, a shallow residual structure is built to share weights, thereby avoiding gradient disappearance or gradient explosion as the network layer deepens; consequently, the difficulty of model optimisation is simplified. Additional convolutional neural networks are used to accelerate the refinement of deep abstract features based on the spatial importance of the gesture feature distribution. Finally, a fully connected cascade softmax classifier is used to complete the gesture recognition. Compared with the dense connection multiplexing feature information network, the proposed algorithm is optimised in feature multiplexing to avoid performance fluctuations caused by feature redundancy. Experimental results from the ISOGD gesture dataset and Gesture dataset prove that the proposed algorithm affords a fast convergence speed and high accuracy.

• Journal of Korea Multimedia Society
• /
• v.22 no.2
• /
• pp.146-156
• /
• 2019

In this paper, we propose a character floating hologram system using the user's gender and age. The proposed system recognizes the gender and age of the user through depth images and color images. The depth images are used to find and normalize facial position. Next, by using facial color images, the age and gender are estimated through an verified database-based model of CNN. Finally, the estimated age and gender are expressed to a character for the floating hologram. The proposed system can be used in a variety of areas, including marketing, advertising, and exhibition events using gender or age.

• Smart Media Journal
• /
• v.7 no.4
• /
• pp.61-69
• /
• 2018

Multi-tracking of general objects and specific faces is an important topic in the field of computer vision applicable to many branches of industry such as biometrics, security, etc. The rapid development of deep neural networks has resulted in a dramatic improvement in face recognition and object detection problems, which helps improve the multiple-face tracking techniques exploiting the tracking-by-detection method. Our proposed method uses face detection trained with a head dataset to resolve the face deformation problem in the tracking process. Further, we use robust face features extracted from the deep face recognition network to match the tracklets with tracking faces using Hungarian matching method. We achieved promising results regarding the usage of deep face features and head detection in a face tracking benchmark.

• Proceedings of the Korean Society of Computer Information Conference
• /
• 2019.01a
• /
• pp.123-124
• /
• 2019

This paper proposes an improved deep learning method based on small data sets for animal image classification. Firstly, we use a CNN to build a training model for small data sets, and use data augmentation to expand the data samples of the training set. Secondly, using the pre-trained network on large-scale datasets, such as VGG16, the bottleneck features in the small dataset are extracted and to be stored in two NumPy files as new training datasets and test datasets. Finally, training a fully connected network with the new datasets. In this paper, we use Kaggle famous Dogs vs Cats dataset as the experimental dataset, which is a two-category classification dataset.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.12 no.12
• /
• pp.6018-6033
• /
• 2018

Fires are a common cause of catastrophic personal injuries and devastating property damage. Every year, many fires occur and threaten human lives and property around the world. Providing early important sign for early fire detection, and therefore the detection of smoke is always the first step in fire-alarm systems. In this paper we propose an automatic smoke detection system built on camera surveillance and image processing technologies. The key features used in our algorithm are to detect and track smoke as moving objects and distinguish smoke from non-smoke objects using a convolutional neural network (CNN) model for cascade classification. The results of our experiment, in comparison with those of some earlier studies, show that the proposed algorithm is very effective not only in detecting smoke, but also in reducing false positives.

• International journal of advanced smart convergence
• /
• v.10 no.3
• /
• pp.163-171
• /
• 2021

Recently, Machine Learning-based visualization approaches have been proposed to combat the problem of malware detection. Unfortunately, these techniques are exposed to Adversarial examples. Adversarial examples are noises which can deceive the deep learning based malware detection network such that the malware becomes unrecognizable. To address the shortcomings of these approaches, we present Block-matching and 3D filtering (BM3D) algorithm and deep image prior based denoising technique to defend against adversarial examples on visualization-based malware detection systems. The BM3D based denoising method eliminates most of the adversarial noise. After that the deep image prior based denoising removes the remaining subtle noise. Experimental results on the MS BIG malware dataset and benign samples show that the proposed denoising based defense recovers the performance of the adversarial attacked CNN model for malware detection to some extent.

• Journal of Korea Multimedia Society
• /
• v.24 no.8
• /
• pp.988-999
• /
• 2021

AOI (Automatic Optical Inspection) of PCB (Printed Circuit Board) is a very important step to guarantee the product performance. The process of registering components called teaching mode is first perform, and AOI is then carried out in a testing mode that checks defects, such as recognizing and comparing the component mounted on the PCB to the stored components. Since most of registration of the components on the PCB is done manually, it takes a lot of time and there are many problems caused by mistakes or misjudgement. In this paper, A components classifier is proposed using YOLO (You Only Look Once) v2's object detection model that can automatically register components in teaching modes to reduce dramatically time and mistakes. The network of YOLO is modified to classify small objects, and the number of anchor boxes was increased from 9 to 15 to classify various types and sizes. Experimental results show that the proposed method has a good performance with 99.86% accuracy.