• Journal of Korea Multimedia Society
• /
• v.14 no.6
• /
• pp.711-718
• /
• 2011

The k-Nearest Neighbor classifier that does not require a training phase is appropriate for a variable number of classes problem like face recognition, Recently distance metric learning methods that is trained with a given data set have reported the significant improvement of the kNN classifier. However, the performance of a distance metric learning method is variable for each application, In this paper, we focus on the face recognition and compare the performance of the state-of-the-art distance metric learning methods, Our experimental results on the public face databases demonstrate that the Mahalanobis distance metric based on PCA is still competitive with respect to both performance and time complexity in face recognition.

• Journal of Korea Multimedia Society
• /
• v.23 no.6
• /
• pp.729-737
• /
• 2020

Screening with low-dose spiral computed tomography (LDCT) has been shown to reduce lung cancer mortality by about 20% when compared to standard chest radiography. One of the problems arising from screening programs is that large amounts of CT image data must be interpreted by radiologists. To solve this problem, automated detection of pulmonary nodules is necessary; however, this is a challenging task because of the high number of false positive results. Here we demonstrate detection of pulmonary nodules using six off-the-shelf convolutional neural network (CNN) models after modification of the input/output layers and end-to-end training based on publicly databases for comparative evaluation. We used the well-known CNN models, LeNet-5, VGG-16, GoogLeNet Inception V3, ResNet-152, DensNet-201, and NASNet. Most of the CNN models provided superior results to those of obtained using customized CNN models. It is more desirable to modify the proven off-the-shelf network model than to customize the network model to detect the pulmonary nodules.

• Journal of the Korea Computer Industry Society
• /
• v.2 no.8
• /
• pp.1117-1122
• /
• 2001

Recently, issues on the internet searching of image information through various multimedia databases have drawn an tremendous attention and several researches on image information retrieval methods are on progress. By incorporating wavelet transform and correlation matrixes, we propose a novel and highly efficient feature vector extraction algorithm that has an capability of a robust similarity matching. The simulation results have yielded a faster and highly accurate candidate image retrieval performance in comparison to those of the conventional algorithms. Such an improved performance can be obtained because the used feature vectors were compressed to 256：1 while the correlation matrixes are incorporated to provide a fuel information for the better matching.

• Journal of Korea Multimedia Society
• /
• v.9 no.12
• /
• pp.1552-1559
• /
• 2006

In a web environment, similar information exists in many different places in diverse formats. Even duplicate information is stored in the various databases using different terminologies. Since most information serviced in the current World Wide Web however had been constructed before the advent of ontology, it is practically almost impossible to construct ontology for all those resources in the web. In this paper, we assume that most information in the web environment exist in the form of RDBMS, and propose a matching method between domain ontology and existing RDBMS tables for semantic retrieval. In the processing of extracting a local ontology, some problems such as losing domain in formation can occur since the correlation of domain ontology has not been considered at all. To prevent these problems, we propose an instance-based matching which uses relational information between RDBMS tables and relational information between classes in domain ontology. To verify the efficiency of the method proposed in this paper, several experiments are conducted using the digital heritage information currently serviced in the countrywide museums. Results show that the proposed method increase retrieval accuracy in terms of user relevance and satisfaction.

• Journal of Korea Multimedia Society
• /
• v.12 no.10
• /
• pp.1450-1456
• /
• 2009

By the explosive growth of IT technologies, mobile phones have embedded a lot of functions and everyone can use them with facility. But there are various cybercrimes as invasions of one's privacy or thefts of company's sensitive information using a built-in digital camera function in a mobile phone. In this paper, we propose a scheme for analyzing evidences by digital pictures on mobile phones. Therefore we analyze the features of digital pictures on mobile phones and make databases of characteristic patterns based on the vendor and the model of mobile phone. The proposed scheme will help to acquire digital evidences by providing a better decision of the vendor and/or the model of mobile phone by cybercrime suspects.

• Journal of Korea Multimedia Society
• /
• v.16 no.10
• /
• pp.1156-1162
• /
• 2013

Face verification has been widely studied during the past two decades. One of the challenges is the rising concern about the security and privacy of the template database. In this paper, we propose a secure face verification system which generates a unique secure cryptographic key from a face template. The face images are processed to produce face templates or codes to be utilized for the encryption and decryption tasks. The result identity data is encrypted using Advanced Encryption Standard (AES). Distance metric naming hamming distance and Euclidean distance are used for template matching identification process, where template matching is a process used in pattern recognition. The proposed system is tested on the ORL, YALEs, and PKNU face databases, which contain 360, 135, and 54 training images respectively. We employ Principle Component Analysis (PCA) to determine the most discriminating features among face images. The experimental results showed that the proposed distance measure was one the promising best measures with respect to different characteristics of the biometric systems. Using the proposed method we needed to extract fewer images in order to achieve 100% cumulative recognition than using any other tested distance measure.

• The Transactions of the Korea Information Processing Society
• /
• v.4 no.12
• /
• pp.2931-2946
• /
• 1997

In multimedia information retrieval applications, content-based image retrieval is essential for retrieving relevant multimedia documents. The purpose of our paper is to provide effective representation and efficient retrieval of images when a pixel-level original image is automatically or manually transformaed into its iconic image containing meaningful graphic descriptions, called icon objects. For this, we first propose new spatial match representationschemes to describe spatial relationships between icon objects accurately by expressing them as rectangles, rather than as points. In order to accelerate image searching, we also design an efficient retrieval method using a two-dimensional signature file organization. Finally, we show from our experiment that the proposed representation schemes achieve better retrieval effectiveness than the 9-DLT (Direction Lower Triangular) scheme.

• Journal of Korea Multimedia Society
• /
• v.22 no.6
• /
• pp.655-664
• /
• 2019

In this paper, we propose a novel way of combining multiple deep convolutional neural network (DCNN) architectures which work well for accurate video face identification by adopting a serial combination of 3D and 2D DCNNs. The proposed method first divides an input video sequence (to be recognized) into a number of sub-video sequences. The resulting sub-video sequences are used as input to the 3D DCNN so as to obtain the class-confidence scores for a given input video sequence by considering both temporal and spatial face feature characteristics of input video sequence. The class-confidence scores obtained from corresponding sub-video sequences is combined by forming our proposed class-confidence matrix. The resulting class-confidence matrix is then used as an input for learning 2D DCNN learning which is serially linked to 3D DCNN. Finally, fine-tuned, serially combined DCNN framework is applied for recognizing the identity present in a given test video sequence. To verify the effectiveness of our proposed method, extensive and comparative experiments have been conducted to evaluate our method on COX face databases with their standard face identification protocols. Experimental results showed that our method can achieve better or comparable identification rate compared to other state-of-the-art video FR methods.

• Journal of Korea Multimedia Society
• /
• v.22 no.3
• /
• pp.366-373
• /
• 2019

The development of monitoring and analysis systems to increase productivity while saving energy is needed as a method to reduce huge amount of energy consumed in the process of producing large forged products. In this paper, we propose a system to monitor and analyze energy usage in real-time collected from gas-meter, wattmeter, and thermometer based on IoT installed in forging factories. The system consists of a data collection server for collecting and processing data from IoT- based platform and existing SCADA equipment and ERP/MES system in forging factories, and an application server for providing services to users. To develop the system, the overall system structure is logically diagrammed, and the databases configuration and implementation modules to efficiently store and manage data are presented. In the future, the system will be utilized to reduce energy consumption by analyzing energy usage pattern and optimizing process works with real-time energy usage and production process data for each facility.

• Journal of KIISE:Databases
• /
• v.29 no.1
• /
• pp.15-26
• /
• 2002

Most work on multimedia storage systems has assumed that client will be serviced using a round-robin strategy. The server services the clients in rounds and each client is allocated a time slice within that round. Furthermore, most such algorithms are evaluated on the basis of a tightly coupled cost function. This is the basis of well-known algorithm such as FCFS, SCAN, SCAN-EDF, etc. In this paper, we describe a scheduling module called Request Unifier(RU) that takes as input, a set of client request, and a set of constraints on the desired performance such as client waiting time or maximum disk bandwidth, and a cost function. It produces as output a Unified Read Request(URR), telling the storage server which data items to read and when these data items to be delivered to the clients. Given a cost function, a URR is optimal if there is no other URR satisfying the constraints with a lower cost. We present three algorithms in this paper that can accomplish this kind of request merging and compare their performance through an experimental evaluation.