• The Journal of the Acoustical Society of Korea
• /
• v.19 no.2
• /
• pp.40-45
• /
• 2000

In speech corpus generation and speech recognition, it is sometimes needed to segment the input speech data without any prior knowledge. A method to accomplish this kind of segmentation, often called as blind segmentation, or acoustic segmentation, is to find boundaries which minimize the Euclidean distances among the feature vectors of each segments. However, the use of this metric alone is prone to errors because of the fluctuations or variations of the feature vectors within a segment. In this paper, we introduce the principal component analysis method to take the trend of feature vectors into consideration, so that the proposed distance measure be the distance between feature vectors and their projected points on the principal components. The proposed distance measure is applied in the LBDP(level building dynamic programming) algorithm for an experimentation of continuous speech segmentation. The result was rather promising, resulting in 3-6% reduction in deletion rate compared to the pure Euclidean measure.

• The Journal of Korean Institute of Information Technology
• /
• v.17 no.5
• /
• pp.13-20
• /
• 2019

Recently, object detection technology based on CNN has been actively studied. Object detection technology is used as an important technology in autonomous vehicles, intelligent image analysis, and so on. In this paper, we propose a rotation change robust object detection system by applying MI-FL (Moment Invariant-Feature Layer) to SSD (Single Shot Multibox Detector) which is one of CNN-based object detectors. First, the features of the input image are extracted based on the VGG network. Then, a total of six feature layers are applied to generate bounding boxes by predicting the location and type of object. We then use the NMS algorithm to get the bounding box that is the most likely object. Once an object bounding box has been determined, the invariant moment feature of the corresponding region is extracted using MI-FL, and stored and learned in advance. In the detection process, it is possible to detect the rotated image more robust than the conventional method by using the previously stored moment invariant feature information. The performance improvement of about 4 ~ 5% was confirmed by comparing SSD with existing SSD and MI-FL.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2014.11a
• /
• pp.242-244
• /
• 2014

파노라마 영상은 하나의 영상이 가지는 제한된 시점의 한계를 극복하고 폭넓은 시야를 가질 수 있다는 점에서 최근 여러 분야에서 활용되고 있는 기술이다. 본 논문에서는 자연스러운 파노라마 영상 생성을 위해 SURF(speed up robust feature)를 이용한 특징점 기반의 파노라마 영상 생성 기법을 제안한다. SURF 알고리즘을 사용하면 정합할 두 영상에서 특징점들을 추출할 수 있다. 추출된 특징점들을 RANSAC(random sample consensus) 알고리즘을 통해 특징점 간 정합시 오차율을 최소화한다. 또한, 이미지 왜곡을 최소화하기 위해 실린더 투영을 이용하여 영상을 보정한다. 최종적으로, 서로 다른 두 영상을 합성할 때 발생하는 경계 주변의 이질감을 보완하기 위해 블렌딩 기법을 사용함으로써 자연스러운 파노라마 영상을 생성한다.

• Proceedings of the Korean Information Science Society Conference
• /
• 2006.06a
• /
• pp.133-135
• /
• 2006

단층촬영영상(Tomographic cross-section images)으로부터 임의의 등밀도 표면(iso-density surface)을 재구성하기 위한 새로운 방법을 제안하였다. 이 방법에서는 마칭큐브 알고리즘에 비해 정밀도는 떨어지지만 안정적인 표면을 생성하는 셀경계 알고리즘(Cell-Boundary Method)을 이용하여 초기메쉬를 구하고 이를 표면축소포장(Shrink-wrapping)처리를 통해 정밀한 등밀도 표면을 생성하게 된다. 이는 마칭큐브와 같이 단층영상에서 등밀도 표면을 직접 추출하는 것이 아니라 등밀도점(iso-density Point)을 먼저 추출하고 표면의 모호성이 없는 안정적인 초기메쉬를 이들 방향으로 축소하여 정확한 표면모델링을 가능하게 한다. 이를 통해 마칭큐브에서 발생하는 표면 결정의 모호성이 없이 보다 안정적인 표면을 정확하게 만들 수 있다.

• Korean Journal of Remote Sensing
• /
• v.36 no.2_2
• /
• pp.337-349
• /
• 2020

In this study, Helikite, Low Altitude Remote Sensing (LARS) platform, was used to acquire coastal images. In the obtained image, the land and water masses were divided using four types of region clustering algorithms, and then waterline was extracted using edge detection. Quantitative comparisons were not possible due to the lack of in-situ waterline data. But, based on the image of the infrared band where water masses and land are relatively clear, the waterlines extracted by each algorithm were compared. As a result, it was found that each algorithm differed significantly in the part where the distinction between water masses and land was ambiguous. This is considered to be a difference in the process of selecting the threshold value of the digital number that each algorithm uses to distinguish the regions. The extraction of waterlines through various algorithms is expected to be used in conjunction with a Low Altitude Remote Sensing system that can be continuously monitored in the future to explain the rapid changes in coastal shape through several years of long-term data from fixed areas.

• Korean Journal of Remote Sensing
• /
• v.20 no.4
• /
• pp.261-274
• /
• 2004

The feature extraction using remotely sensed imagery has been recognized one of the important tasks in remote sensing applications. As the high-resolution imagery are widely used to the engineering purposes, need of more accurate feature information also is increasing. Especially, in case of the automatic extraction of linear feature such as road using mid or low-resolution imagery, several techniques was developed and applied in the mean time. But quantitatively comparative analysis of techniques and case studies for high-resolution imagery is rare. In this study, we implemented a computer program to perform and compare GDPA (Gradient Direction Profile Analysis) algorithm and Hough transformation. Also the results of applying two techniques to some images were compared with road centerline layers and boundary layers of digital map and presented. For quantitative comparison, the ranking method using commission error and omission error was used. As results, Hough transform had high accuracy over 20% on the average. As for execution speed, GDPA shows main advantage over Hough transform. But the accuracy was not remarkable difference between GDPA and Hough transform, when the noise removal was app]ied to the result of GDPA. In conclusion, it is expected that GDPA have more advantage than Hough transform in the application side.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.21 no.10
• /
• pp.1892-1897
• /
• 2017

This paper proposes a finger counting algorithm for a hand with stuck fingers. The proposed algorithm is based on the fact that straight line type shadows are inevitably generated between fingers. It divides the hand region into the thumb region and the four fingers region for effective shadow detection, and generates an edge image in each region. Projection curves are generated by appling a line detection and a projection technique to each edge image, and the peaks of the curves are detected as candidates for finger shadows. And then peaks due to finger shadows are extracted from them and counted. In the finger counting experiment on hand images expressing various shapes with stuck fingers, the counting success rate is from 83.3% to 100% according to the number of fingers, and 93.1% on the whole. It also shows that if hand images are generated under controlled conditions, the failure cases can be sufficiently improved.

• 한국HCI학회:학술대회논문집
• /
• 2008.02c
• /
• pp.19-24
• /
• 2008

This paper presents a physically-based haptic rendering algorithm for a deformable object based on visual information about the intervention between a tool and a real object in a remote place. The physically-based model of a deformable object is created from the mechanical properties of the object and the captured image obtained with a CCD camera. When a slave system exerts manipulation tasks on a deformable object, the reaction force for haptic rendering is computed using boundary element method. Snakes algorithm is used to obtain the geometry information of a deformable object. The proposed haptic rendering algorithm can provide haptic feedback to a user without using a force transducer in a teleoperation system.

• The Journal of the Korea Contents Association
• /
• v.9 no.2
• /
• pp.43-50
• /
• 2009

We proposed similarity detection method of the video frame data that extracts the feature data of own video frame and creates the 1-D signal in this paper. We get the similar frame boundary and make the representative frames within the frame boundary to extract the similarity extraction between video. Representative frames make blurring frames and extract the feature data using DOG values. Finally, we convert the feature data into the 1-D signal and compare the contents similarity. The experimental results show that the proposed algorithm get over 0.9 similarity value against noise addition, rotation change, size change, frame delete, frame cutting.

• Korean Journal of Cognitive Science
• /
• v.14 no.4
• /
• pp.33-42
• /
• 2003

This paper presents a new system for facial expression recognition based in dimension model of internal states. The information of facial expression are extracted to the three steps. In the first step, Gabor wavelet representation extracts the edges of face components. In the second step, sparse features of facial expressions are extracted using fuzzy C-means(FCM) clustering algorithm on neutral faces, and in the third step, are extracted using the Dynamic Model(DM) on the expression images. Finally, we show the recognition of facial expression based on the dimension model of internal states using a multi-layer perceptron. The two dimensional structure of emotion shows that it is possible to recognize not only facial expressions related to basic emotions but also expressions of various emotion.