• Proceedings of the Korea Information Processing Society Conference
• /
• 2023.11a
• /
• pp.691-694
• /
• 2023

본 논문은 CCTV 화면에서의 다양한 이상상황 중 교량 데이터에 특화된 자세 추정 기반 이상탐지 알고리즘을 소개한다. 교량은 크게 도로, 인도 이렇게 두 구역으로 나눠지며, 사람들의 이동방향이 한정적이라는 특징을 가지는 장소 중 하나이다. 이러한 장소적 특징을 이용하고자 사람 자세 추정을 통해 이상의 기준을 잡고 교량 데이터에 특화된 이상탐지 알고리즘을 제안한다. CCTV 영상은 이상을 정하기 어렵고 이상에 대한 레이블이 없는 데이터가 대부분이며 이상에 대한 레이블 생성시 많은 비용 발생이 필수적이다. 본 연구에서는 이러한 한계점을 극복하고자 영상 데이터를 이미지 단위가 아닌 영상 단위로 레이블이 담긴 weakly label 을 가지는 데이터를 활용한 이상탐지 모델을 이용하였다. 특히, 교량에서의 이상상황의 특징인 사람 자세 추정으로 추출한 특질을 추가하여 기존 알고리즘의 이상탐지 예측 성능을 개선하였다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2024.05a
• /
• pp.283-284
• /
• 2024

디지털 워터마킹은 디지털 컨텐츠에 정보를 삽입하는 기술이다. 종래의 디지털 워터마킹 기술은 견고성과 비가시성 사이에 트레이드오프 관계를 가지고, 변형 및 노이즈 공격 등에 취약하다. 본 논문에서는 호스트 이미지의 비가시성을 보장하면서 효율적인 공격 탐지와 소유자 식별이 가능한 워터마킹 기법을 제안한다. 제안한 방식은 주파수 분할 기반의 계층적 워터마킹 및 공격 탐지 시그니처 삽입을 통해 비가시성을 보장하며 용량과 견고성 측면에서 종래의 방법보다 향상된 성능을 보였다. 실험 결과에 따르면 종래의 디지털 워터마크가 무력화되는 왜곡 공격 상황에서 공격 탐지 시그니처 검출이 가능하여 워터마크 공격을 탐지하고 소유자를 식별할 수 있었다.

• KIPS Transactions on Computer and Communication Systems
• /
• v.12 no.12
• /
• pp.363-370
• /
• 2023

Object detection technology that accurately recognizes the road and surrounding conditions is a key technology in the field of autonomous driving. In the field of autonomous driving, object detection technology requires real-time performance as well as accuracy of inference services. Task offloading technology should be utilized to apply object detection technology for accuracy and real-time on resource-constrained devices rather than high-performance machines. In this paper, experiments such as performance comparison of task offloading, performance comparison according to input image resolution, and performance comparison according to camera object resolution were conducted and the results were analyzed in relation to the application of task offloading for real-time object detection of autonomous driving in resource-constrained devices. In this experiment, the low-resolution image could derive performance improvement through the application of the task offloading structure, which met the real-time requirements of autonomous driving. The high-resolution image did not meet the real-time requirements for autonomous driving due to the increase in communication time, although there was an improvement in performance. Through these experiments, it was confirmed that object recognition in autonomous driving affects various conditions such as input images and communication environments along with the object recognition model used.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.21 no.5
• /
• pp.451-457
• /
• 2020

Format recognition and OCR techniques are widely used as methods for detecting and protecting personal information from electronic documents. However, due to the poor recognition rate of the OCR engine, personal information cannot be detected or false positives commonly occur. It also takes a long time to analyze a large amount of electronic documents. In this paper, we propose a method to improve the speed of image analysis of electronic documents, character recognition rate of the OCR engine, and detection rate of personal information by improving the existing method. The analysis speed was increased using the format recognition method while the analysis speed and character recognition rate of the OCR engine was improved by image correction. An algorithm for analyzing personal information from images was proposed to increase the reconnaissance rate of personal information. Through the experiments, 1755 image format recognition samples were analyzed in an average time of 0.24 seconds, which was 0.5 seconds higher than the conventional PAID system format recognition method, and the image recognition rate was 99%. The proposed method in this paper can be used in various fields such as public, telecommunications, finance, tourism, and security as a system to protect personal information in electronic documents.

• Journal of Digital Contents Society
• /
• v.19 no.6
• /
• pp.1177-1183
• /
• 2018

Recent developments in machine learning have attracted a lot of attention for techniques such as machine learning and deep learning that implement artificial intelligence. In this paper, binary malicious code using deep learning based R-CNN is imaged and the feature is extracted from the image to classify the family. In this paper, two steps are used in deep learning to image malicious code using CNN. And classify the characteristics of the family of malicious codes using R-CNN. Generate malicious code as an image, extract features, classify the family, and automatically classify the evolution of malicious code. The detection rate of the proposed method is 93.4% and the accuracy is 98.6%. In addition, the CNN processing speed for image processing of malicious code is 23.3 ms, and the R-CNN processing speed is 4ms to classify one sample.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2022.05a
• /
• pp.667-669
• /
• 2022

2차원 이미지를 통한 자세 추정의 경우 관절이 겹치거나 가려져 있는 등의 인식 저해 요소로 인하여 자세 추정 정확도가 감소하는 한계가 있다. 본 논문에서는 GAN을 통해 2차원 이미지를 3차원으로 증강한 뒤 자세를 추정하는 기법을 제안한다. 제안하는 방법은 2차원 이미지의 평면좌표 값에서 GAN을 통해 노이즈 벡터 z축 값과 피사체에 투영되는 빛의 방향 값을 반영한 3차원 이미지를 만든다. 이러한 이미지 합성 과정을 거친 후 DeepLabCut을 사용해 관절 좌표를 추출하고 자세 추정 및 분류를 진행한다. 이를 통해 2차원에서의 자세 추정 정확도 향상을 기대할 수 있으며, 향후 이를 기반한 이상행동 탐지 분야에서 적용할 수 있다.

• Journal of KIISE:Computing Practices and Letters
• /
• v.15 no.12
• /
• pp.1003-1007
• /
• 2009

Client-side attacks including drive-by download target vulnerabilities in client applications that interact with a malicious server or process malicious data. A typical client-side attack is web-based one related to a malicious web page exploiting specific browser vulnerability that can execute mal ware on the client system (PC) or give complete control of it to the malicious server. To defend those attacks, this paper has constructed high interaction client honeypot system using Capture-HPC that adopts execution-based detection in virtual machine. We have detected and classified malicious web pages using the system. We have also analyzed the system's performance in terms of the number of virtual machine images and the number of browsers executed simultaneously in each virtual machine. Experimental results show that the system with one virtual machine image obtains better performance with less reverting overhead. The system also shows good performance when the number of browsers executed simultaneously in a virtual machine is 50.

• Convergence Security Journal
• /
• v.22 no.5
• /
• pp.11-20
• /
• 2022

As the Internet and web technology develop around mobile devices, image data contains various types of sensitive information such as people, text, and space. In addition to these characteristics, as the use of SNS increases, the amount of damage caused by exposure and abuse of personal information online is increasing. However, research on de-identification technology based on multi-type object detection for personal information protection is insufficient. Therefore, this paper proposes an artificial intelligence model that detects and de-identifies multiple types of objects using existing single-type object detection models in parallel. Through cutmix, an image in which person and text objects exist together are created and composed of training data, and detection and de-identification of objects with different characteristics of person and text was performed. The proposed model achieves a precision of 0.724 and mAP@.5 of 0.745 when two objects are present at the same time. In addition, after de-identification, mAP@.5 was 0.224 for all objects, showing a decrease of 0.4 or more.

• Journal of IKEEE
• /
• v.27 no.3
• /
• pp.280-287
• /
• 2023

Deep learning advancements can solve computer vision problems, but large-scale datasets are necessary for high accuracy. In this paper, we propose an image generation technique using object bounding boxes and image edge components. The object bounding boxes are extracted from the images through object detection, and image edge components are used as input values for the image generation model to create new image data. As results of experiments, the images generated by the proposed method demonstrated similar image quality to the source images in the image quality assessment, and also exhibited good performance during the deep learning training process.

• Convergence Security Journal
• /
• v.7 no.4
• /
• pp.23-28
• /
• 2007

In this research, we suggest the unknown intrusion detection system with SVM(Support Vector Machines). At the system, at first, collected training-packets are processed through packet image creating module. And then, it is studied by the SVM module. Finally, the studied SVM module classifies the test-data unsing test-packet-image. This system's stability and efficient characteristic of security is far superior than the existing it.