• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.23 no.1
• /
• pp.84-89
• /
• 2019

JPEG is a most widely used standard image compression technology. This research introduces the JPEG image compression algorithm and describes each step in the compression and decompression. Image compression is the application of data compression on digital images. The DCT (discrete cosine transform) is a technique for converting a time domain to a frequency domain. First, the image is divided into 8 by 8 pixel blocks. Second, working from top to bottom left to right, the DCT is applied to each block. Third, each block is compressed through quantization. Fourth, the matrix of compressed blocks that make up the image is stored in a greatly reduced amount of space. Finally if desired, the image is reconstructed through decompression, a process using IDCT (inverse discrete cosine transform). The purpose of this research is to review all the processes of image compression / decompression using the discrete cosine transform method.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.26 no.9
• /
• pp.1330-1339
• /
• 2022

3D object detection generally aims to detect relatively large data such as automobiles, buses, persons, furniture, etc, so it is vulnerable to small object detection. In addition, in an environment with limited resources such as embedded devices, it is difficult to apply the model because of the huge amount of computation. In this paper, the accuracy of small object detection was improved by focusing on local features using only one layer, and the inference speed was improved through the proposed knowledge distillation method from large pre-trained network to small network and adaptive quantization method according to the parameter size. The proposed model was evaluated using SUN RGB-D Val and self-made apple tree data set. Finally, it achieved the accuracy performance of 62.04% at mAP@0.25 and 47.1% at mAP@0.5, and the inference speed was 120.5 scenes per sec, showing a fast real-time processing speed.

• Convergence Security Journal
• /
• v.24 no.2
• /
• pp.9-17
• /
• 2024

This paper examines the security threats to enterprise Generative Artificial Intelligence systems and proposes countermeasures. As AI systems handle vast amounts of data to gain a competitive edge, security threats targeting AI systems are rapidly increasing. Since AI security threats have distinct characteristics compared to traditional human-oriented cybersecurity threats, establishing an AI-specific response system is urgent. This study analyzes the importance of AI system security, identifies key threat factors, and suggests technical and managerial countermeasures. Firstly, it proposes strengthening the security of IT infrastructure where AI systems operate and enhancing AI model robustness by utilizing defensive techniques such as adversarial learning and model quantization. Additionally, it presents an AI security system design that detects anomalies in AI query-response processes to identify insider threats. Furthermore, it emphasizes the establishment of change control and audit frameworks to prevent AI model leakage by adopting the cyber kill chain concept. As AI technology evolves rapidly, by focusing on AI model and data security, insider threat detection, and professional workforce development, companies can improve their digital competitiveness through secure and reliable AI utilization.

• Phonetics and Speech Sciences
• /
• v.6 no.1
• /
• pp.77-83
• /
• 2014

This paper proposes a speaker and environment clustering method in order to overcome the degradation of the speech recognition performance caused by various noise and speaker characteristics. In this paper, instead of using the distance between Gaussian mixture model (GMM) weight vectors as in the Google's approach, the distance between the adapted mean vectors based on the modified maximum a posteriori (MAP) adaptation is used as a distance measure for vector quantization (VQ) clustering. According to our experiments on the simulation data generated by adding noise to clean speech, the proposed clustering method yields error rate reduction of 10.6% compared with baseline speaker-independent (SI) model, which is slightly better performance than the Google's approach.

• Journal of information and communication convergence engineering
• /
• v.15 no.1
• /
• pp.7-13
• /
• 2017

We consider a distributed estimation where many nodes remotely placed at known locations collect the measurements of the parameter of interest, quantize these measurements, and transmit the quantized data to a fusion node; this fusion node performs the parameter estimation. Noting that quantizers at nodes should operate in a non-regular framework where multiple codewords or quantization partitions can be mapped from a single measurement to improve the system performance, we propose a low-weight estimation algorithm that finds the most feasible combination of codewords. This combination is found by computing the weighted sum of the possible combinations whose weights are obtained by counting their occurrence in a learning process. Otherwise, tremendous complexity will be inevitable due to multiple codewords or partitions interpreted from non-regular quantized data. We conduct extensive experiments to demonstrate that the proposed algorithm provides a statistically significant performance gain with low complexity as compared to typical estimation techniques.

• The Transactions of the Korean Institute of Electrical Engineers D
• /
• v.53 no.8
• /
• pp.578-584
• /
• 2004

The CSVM(Current Support Vector Machine) that is a digital architecture performing all phases of recognition process including kernel computing, learning, and recall of SVM(Support Vector Machine) on a chip is proposed. Concurrent operation by parallel architecture of elements generates high speed and throughput. The classification problems of bio data having high dimension are solved fast and easily using the CSVM. Quadratic programming in original SVM learning algorithm is not suitable for hardware implementation, due to its complexity and large memory consumption. Hardware-friendly SVM learning algorithms, kernel adatron and kernel perceptron, are embedded on a chip. Experiments on fixed-point algorithm having quantization error are performed and their results are compared with floating-point algorithm. CSVM implemented on FPGA chip generates fast and accurate results on high dimensional cancer data.

• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.26 no.3
• /
• pp.227-239
• /
• 2008

Processing LiDAR (Light Detection And Ranging) data obtained from ALS (Airborne Laser Scanning) systems mainly involves organization and segmentation of the data for 3D object modeling and mapping purposes. The ALS systems are viable and becoming more mature technology in various applications. ALS technology requires complex integration of optics, opto-mechanics and electronics in the multi-sensor components, Le. data captured from GPS, INS and laser scanner. In this study, digital image processing techniques mainly were implemented to gray level coded image of the LiDAR data for building extraction and superstructures segmentation. One of the advantages to use gray level image is easy to apply various existing digital image processing algorithms. Gridding and quantization of the raw LiDAR data into limited gray level might introduce smoothing effect and loss of the detail information. However, smoothed surface data that are more suitable for surface patch segmentation and modeling could be obtained by the quantization of the height values. The building boundaries were precisely extracted by the robust edge detection operator and regularized with shape constraints. As for segmentation of the roof structures, basically region growing based and gap filling segmentation methods were implemented. The results present that various image processing methods are applicable to extract buildings and to segment surface patches of the superstructures on the roofs. Finally, conceptual methodology for extracting characteristic information to reconstruct roof shapes was proposed. Statistical and geometric properties were utilized to segment and model superstructures. The simulation results show that segmentation of the roof surface patches and modeling were possible with the proposed method.

• Journal of Broadcast Engineering
• /
• v.22 no.3
• /
• pp.295-312
• /
• 2017

Digital video content is the most information-intensive and high-value content. Therefore, it is necessary to protect the intellectual property rights of these contents, and this paper also proposes a watermarking method of digital image for this purpose. The proposed method uses the frequency characteristics of 2-Dimensional Discrete Wavelet Transform (2D-DWT) for digital images and digital watermark on global data without using local or specific data of the image for watermark embedding. The method to insert digital watermark data uses a simple Quantization Index Modulation (QIM) and a multiple watermarking method that inserts the same watermark data in multiple. When extracting a watermark, multiple watermarks are extracted and the final watermark data is determined by a simple statistical method. This method is an empirical method for experimentally determining the parameters in the watermark embedding process. The proposed method performs experiments on various images against various attacks and shows the superiority of the proposed method by comparing the performance with the representative existing methods.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.14 no.1
• /
• pp.323-343
• /
• 2020

With the rapid development of computer and communication techniques, copyright protection of vector geographic data has attracted considerable research attention because of the high cost of such data. A novel adaptive watermark detection algorithm is proposed for vector geographic data that can be used to qualitatively analyze the robustness of watermarks against data addition attacks. First, a watermark was embedded into the vertex coordinates based on coordinate mapping and quantization. Second, the adaptive watermark detection model, which is capable of calculating the detection threshold, false positive error (FPE) and false negative error (FNE), was established, and the characteristics of the adaptive watermark detection algorithm were analyzed. Finally, experiments were conducted on several real-world vector maps to show the usability and robustness of the proposed algorithm.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.8 no.7
• /
• pp.2572-2589
• /
• 2014

Reversible data hiding is a technique for recovering original images without any distortion after secret data are extracted from the image. The technique continues to attract attention from many researchers. In this paper, we introduce a new reversible data hiding scheme based on the adjacent index differences of vector quantization (VQ) indices. The proposed scheme exploits the differences between two adjacent indices to embed secret data. Experimental results show that our scheme can achieve a lower compression rate than an earlier scheme by Yang and Lin. Our scheme's average compression rate, 0.44 bpp, outperforms that of Yang and Lin's scheme, which averages 0.53 bpp. Moreover, the embedding capacity of our scheme can rise to 1.45 bpi, which also is superior to that of Chang et al.'s scheme [35] (1.00 bpi)Yang and Lin's scheme [27] (0.91 bpi) as well as Chang et al.'s scheme [26] (0.74 bpi).