• Journal of the Korean Society of Radiology
• /
• v.6 no.1
• /
• pp.63-71
• /
• 2012

This study aimed to evaluate the accuracy of breast cancer diagnosis of digital mammography which is in the highest interest of breast imaging test, and to investigate the characteristics of breast cancer patients. For this purpose, 57 breast cancer patients who underwent breast imaging test were examined between May 2010 and June 2011. The average age of the breast cancer patients was 50.8 years old, and the most frequently occurring location was the upper outer quadrant (UOQ), accounting for 33.3%. By age, the highest occurrence rate of breast cancer was the age group of 40~49, accounting for 42.1%. As for the breast composition of the breast cancer patients, fatty breast accounted for 31.6% (18/57) and dense breast for 68.4% (39/57), indicating that nearly 70% of the breast cancer patients have dense breast. It was found that the detection rate of breast cancer was the highest (45.3%) when both microcalcification and mass are simultaneously present in the radiographic lesion of the breast imaging. In dense breast, the mass without microcalcification was lower in detection rate than fatty breast. Accordingly, the mass is the cause of raising the false negative rate in dense breast. The findings show that the false negative rate of digital mammography was 7.0% and the sensitivity 93.0%. Also, the false negative rate of dense breast was 12.8%, and the sensitivity 87.2%, indicating that the sensitivity to breast cancer in this study was higher than the dense breast of previously reported screen film mammography.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.7 no.11
• /
• pp.2720-2736
• /
• 2013

Multi-view face detection has become an active area for research in the last few years. In this paper, a novel multi-view human face detection algorithm based on improved real Adaboost is presented. Real Adaboost algorithm is improved by weighted combination of weak classifiers and the approximately best combination coefficients are obtained. After that, we proved that the function of sample weight adjusting method and weak classifier training method is to guarantee the independence of weak classifiers. A coarse-to-fine hierarchical face detector combining the high efficiency of Haar feature with pose estimation phase based on our real Adaboost algorithm is proposed. This algorithm reduces training time cost greatly compared with classical real Adaboost algorithm. In addition, it speeds up strong classifier converging and reduces the number of weak classifiers. For frontal face detection, the experiments on MIT+CMU frontal face test set result a 96.4% correct rate with 528 false alarms; for multi-view face in real time test set result a 94.7 % correct rate. The experimental results verified the effectiveness of the proposed approach.

• Journal of Korea Multimedia Society
• /
• v.19 no.12
• /
• pp.1909-1918
• /
• 2016

This paper presents a unified framework for joint Convolutional Neural Network (CNN) based vehicle detection by leveraging multi-spectral image pairs. With the observation that under challenging environments such as night vision and limited light source, vehicle detection in a single color image can be more tractable by using additional far-infrared (FIR) image, we design joint CNN architecture for both RGB and FIR image pairs. We assume that a score map from joint CNN applied to overall image can be considered as confidence of vehicle existence. To deal with various scale ratios of vehicle candidates, multi-scale images are first generated scaling an image according to possible scale ratio of vehicles. The vehicle candidates are then detected on local maximal on each score maps. The generation of overlapped candidates is prevented with non-maximal suppression on multi-scale score maps. The experimental results show that our framework have superior performance than conventional methods with a joint framework of multi-spectral image pairs reducing false positive generated by conventional vehicle detection framework using only single color image.

• Journal of Korea Multimedia Society
• /
• v.19 no.12
• /
• pp.1927-1935
• /
• 2016

This paper propose a framework for human body parts in RGB-D image. We conduct tasks of obtaining person area, finding candidate areas and local detection in order to detect hand, foot and head which have features of long accumulative geodesic distance. A person area is obtained with background subtraction and noise removal by using depth image which is robust to illumination change. Finding candidate areas performs construction of graph model which allows us to measure accumulative geodesic distance for the candidates. Instead of raw depth map, our approach constructs graph model with segmented regions by quadtree structure to improve searching time for the candidates. Local detection uses HOG based SVM for each parts, and head is detected for the first time. To minimize false detections for hand and foot parts, the candidates are classified with upper or lower body using the head position and properties of geodesic distance. Then, detect hand and foot with the local detectors. We evaluate our algorithm with datasets collected Kinect v2 sensor, and our approach shows good performance for head, hand and foot detection.

• Journal of KIISE:Software and Applications
• /
• v.37 no.10
• /
• pp.760-767
• /
• 2010

There are many memory leak detection tools. However, programmers, who develop very large programs, tend to avoid testing their programs with memory leak detection tools since these tools require runtime and space overheads. Thus, we present a memory leak detection technique which enables programmers to test their modules in their unit test phase with low overheads. To achieve this goal, we extend the existing cyclic memory allocation technique and evaluate our memory leak detection technique on a tiny DBMS. In our experiments, we find our tool has reasonably low runtime and space overheads and it reports only a small number of false positives.

• Journal of Electrical Engineering and Technology
• /
• v.12 no.4
• /
• pp.1406-1416
• /
• 2017

In complex and large-scale industries, properly designed fault detection and identification (FDI) systems considerably improve safety, reliability and availability of target processes. In thermal power plants (TPPs), generating units operate under very dangerous conditions; system failures can cause severe loss of life and property. In this paper, we propose a bagged auto-associative kernel regression (AAKR)-based FDI approach for steam boilers in TPPs. AAKR estimates new query vectors by online local modeling, and is suitable for TPPs operating under various load levels. By combining the bagging method, more stable and reliable estimations can be achieved, since the effects of random fluctuations decrease because of ensemble averaging. To validate performance, the proposed method and comparison methods (i.e., a clustering-based method and principal component analysis) are applied to failure data due to water wall tube leakage gathered from a 250 MW coal-fired TPP. Experimental results show that the proposed method fulfills reasonable false alarm rates and, at the same time, achieves better fault detection performance than the comparison methods. After performing fault detection, contribution analysis is carried out to identify fault variables; this helps operators to confirm the types of faults and efficiently take preventive actions.

• The Journal of Korea Robotics Society
• /
• v.4 no.4
• /
• pp.257-264
• /
• 2009

This paper proposes a novel face detection method that finds tiny faces located at a long range even with low-resolution input images captured by a mobile robot. The proposed approach can locate extremely small-sized face regions of $12{\times}12$ pixels. We solve a tiny face detection problem by organizing a system that consists of multiple detectors including a mean-shift color tracker, short- and long-rage face detectors, and an omega shape detector. The proposed method adopts the long-range face detector that is well trained enough to detect tiny faces at a long range, and limiting its operation to only within a search region that is automatically determined by the mean-shift color tracker and the omega shape detector. By focusing on limiting the face search region as much as possible, the proposed method can accurately detect tiny faces at a long distance even with a low-resolution image, and decrease false positives sharply. According to the experimental results on realistic databases, the performance of the proposed approach is at a sufficiently practical level for various robot applications such as face recognition of non-cooperative users, human-following, and gesture recognition for long-range interaction.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.12 no.7
• /
• pp.3095-3111
• /
• 2018

Spectrum sensing (SS) is one of the fundamental tasks for cognitive radio. In SS, decisions can be made via comparing the test statistics with a threshold. Conventional adaptive algorithms for SS usually adjust their thresholds according to the radio environment. This paper concentrates on the issue of adaptive SS whose threshold is adjusted based on the Markovian behavior of primary user (PU). Moreover, Bayesian cost is adopted as the performance metric to achieve a trade-off between false alarm and missed detection probabilities. Two novel adaptive algorithms, including Markov Bayesian energy detection (MBED) algorithm and IMBED (improved MBED) algorithm, are proposed. Both algorithms model the behavior of PU as a two-state Markov process, with which their thresholds are adaptively adjusted according to the detection results at previous slots. Compared with the existing Bayesian energy detection (BED) algorithm, MBED algorithm can achieve lower Bayesian cost, especially in high signal-to-noise ratio (SNR) regime. Furthermore, it has the advantage of low computational complexity. IMBED algorithm is proposed to alleviate the side effects of detection errors at previous slots. It can reduce Bayesian cost more significantly and in a wider SNR region. Simulation results are provided to illustrate the effectiveness and efficiencies of both algorithms.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.13 no.10
• /
• pp.4958-4970
• /
• 2019

As technology advances, the amount of fake news is increasing more and more by various reasons such as political issues and advertisement exaggeration. However, there have been very few research works on fake news detection, especially which uses grammatical transformation on deep neural network. In this paper, we shall present a new Fake News Detection Model, called FAGON(Fake news detection model using Grammatical transformation On deep Neural network) which determines efficiently if the proposition is true or not for the given article by learning grammatical transformation on neural network. Especially, our model focuses the Korean language. It consists of two modules: sentence generator and classification. The former generates multiple sentences which have the same meaning as the proposition, but with different grammar by training the grammatical transformation. The latter classifies the proposition as true or false by training with vectors generated from each sentence of the article and the multiple sentences obtained from the former model respectively. We shall show that our model is designed to detect fake news effectively by exploiting various grammatical transformation and proper classification structure.

• Current Optics and Photonics
• /
• v.4 no.3
• /
• pp.210-220
• /
• 2020

Aiming at the problem that the Local Sparse Difference Index algorithm has low accuracy and low efficiency when detecting target anomalies in a hyperspectral image, this paper proposes a Weighted Collaborative Representation and Sparse Difference-Based Hyperspectral Anomaly Detection algorithm, to improve detection accuracy for a hyperspectral image. First, the band subspace is divided according to the band correlation coefficient, which avoids the situation in which there are multiple solutions of the sparse coefficient vector caused by too many bands. Then, the appropriate double-window model is selected, and the background dictionary constructed and weighted according to Euclidean distance, which reduces the influence of mixing anomalous components of the background on the solution of the sparse coefficient vector. Finally, the sparse coefficient vector is solved by the collaborative representation method, and the sparse difference index is calculated to complete the anomaly detection. To prove the effectiveness, the proposed algorithm is compared with the RX, LRX, and LSD algorithms in simulating and analyzing two AVIRIS hyperspectral images. The results show that the proposed algorithm has higher accuracy and a lower false-alarm rate, and yields better results.