• 한국멀티미디어학회논문지
• /
• 제22권9호
• /
• pp.1000-1010
• /
• 2019

Image semantic segmentation and dehazing are key tasks in the computer vision. In recent years, researches in both tasks have achieved substantial improvements in performance with the development of Convolutional Neural Network (CNN). However, most of the previous works for semantic segmentation assume the images are captured in clear weather and show degraded performance under hazy images with low contrast and faded color. Meanwhile, dehazing aims to recover clear image given observed hazy image, which is an ill-posed problem and can be alleviated with additional information about the image. In this work, we propose a deep multi-task network for simultaneous semantic segmentation and dehazing. The proposed network takes single haze image as input and predicts dense semantic segmentation map and clear image. The visual information getting refined during the dehazing process can help the recognition task of semantic segmentation. On the other hand, semantic features obtained during the semantic segmentation process can provide cues for color priors for objects, which can help dehazing process. Experimental results demonstrate the effectiveness of the proposed multi-task approach, showing improved performance compared to the separate networks.

• Computers and Concrete
• /
• 제5권1호
• /
• pp.61-74
• /
• 2008

Digital image processing algorithms for the analysis and characterization of grains and voids in cemented materials were developed using toolbox functions of a mathematical software package. Utilization of grayscale, color and watershed segmentation algorithms and their performances were demonstrated on artificially prepared self-compacting concrete (SCC) samples. It has been found that color segmentation was more advantageous over the gray scale segmentation for the detection of voids whereas the latter method provided satisfying results for the aggregate grains due to the sharp contrast between their colors and the cohesive matrix. The watershed segmentation method, on the other hand, appeared to be very efficient while separating touching objects in digital images.

• 한국음향학회지
• /
• 제25권5호
• /
• pp.213-221
• /
• 2006

합성음의 음질을 향상시키기 위하여 분할된 corpora로부터 합성유닛을 선택하여 사용하는 연속음성합성에서 정확한 음소분할은 매우 중요하다. 일반적으로 음소분할은 사람에 의해 수행되지만 많은 작업량으로 인한 시간적 지연, 일관 성 유지 어려움 등 많은 문제가 발생한다. 이에 따라 음성인식에서 도입된 HMM 기반의 자동음소분할이 음성인식, 음성 합성에서 널리 사용되어지고 있지만 음성전문가의 수작업 결과와 비교할 때 HMM 기반 자동음소분할은 오류가 있고, 이는 합성음 품질의 열화의 주요 원인이 되고 있다. 본 논문에서는 HMM 기반의 자동음소분할기를 사용하여 나타난 자동음소분할 결과와 수작업에 의한 음소분할 결과를 비교하고 유형별로 분석함으로써 음성합성의 성능향상을 위해 개선해야 할 문제점들을 제시한다. 실험에서는 ETRI의 표준형 한국어 공통 음성 DB을 사용하였고, 오차의 범위가 20ms를 벗어난 경우를 분절 오류로 간주하였다. 실험 결과 여성화자의 경우 파열음 + 모음, 파찰음 + 모음, 모음 + 유음 음소쌍에서는 각각 약 99%, 99.5%, 99%의 높은 정확률을 보인 반면, 폐쇄음 + 비음, 폐쇄음 + 유음, 비음 + 유음 음소쌍에서는 44.89%, 50%, 55% 의 낮은 정확률을 보였으며, 남성화자에 대한 실험결과에서도 유사한 경향을 보였다.

• 대한인간공학회지
• /
• 제34권5호
• /
• pp.541-548
• /
• 2015

Objective: We introduce a hand gesture segmentation method using a wrist-worn wearable device which can recognize simple gestures of clenching and unclenching ones' fist. Background: There are many types of smart watches and fitness bands in the markets. And most of them already adopt a gesture interaction to provide ease of use. However, there are many cases in which the malfunction is difficult to distinguish between the user's gesture commands and user's daily life motion. It is needed to develop a simple and clear gesture segmentation method to improve the gesture interaction performance. Method: At first, we defined the gestures of making a fist (start of gesture command) and opening one's fist (end of gesture command) as segmentation gestures to distinguish a gesture. The gestures of clenching and unclenching one's fist are simple and intuitive. And we also designed a single gesture consisting of a set of making a fist, a command gesture, and opening one's fist in order. To detect segmentation gestures at the bottom of the wrist, we used a wrist strap on which an array of infrared sensors (emitters and receivers) were mounted. When a user takes gestures of making a fist and opening one's a fist, this changes the shape of the bottom of the wrist, and simultaneously changes the reflected amount of the infrared light detected by the receiver sensor. Results: An experiment was conducted in order to evaluate gesture segmentation performance. 12 participants took part in the experiment: 10 males, and 2 females with an average age of 38. The recognition rates of the segmentation gestures, clenching and unclenching one's fist, are 99.58% and 100%, respectively. Conclusion: Through the experiment, we have evaluated gesture segmentation performance and its usability. The experimental results show a potential for our suggested segmentation method in the future. Application: The results of this study can be used to develop guidelines to prevent injury in auto workers at mission assembly plants.

• 한국정보처리학회:학술대회논문집
• /
• 한국정보처리학회 2012년도 춘계학술발표대회
• /
• pp.436-439
• /
• 2012

The work is to the control the desktop Computers based on hand gesture recognition. This paper is worked en real time tracking and recognizes the hand gesture for controlling the slides based on hand direction such as right and left using a real time camera.

• 융복합기술연구소 논문집
• /
• 제11권1호
• /
• pp.1-5
• /
• 2021

Deep learning based methods achieve state-of-the-art accuracy, however, they typically rely on supervised training with large labeled datasets. It is known in many medical applications that labeling medical images requires significant expertise and much time, and typical hand-tuned approaches for data augmentation fail to capture the complex variations in such images. This paper proposes a 3D image augmentation method to overcome these difficulties. It allows us to enrich diversity of training data samples that is essential in medical image segmentation tasks, thus reducing the data overfitting problem caused by the fact the scale of medical image dataset is typically smaller. Our numerical experiments demonstrate that the proposed approach provides significant improvements over state-of-the-art methods for 3D medical image segmentation.

• 대한음성학회지:말소리
• /
• 제58호
• /
• pp.67-81
• /
• 2006

For a large corpus of time-aligned data, HMM based approaches are most widely used for automatic segmentation, providing a consistent and accurate phone labeling scheme. There are two methods for training in HMM. Flat starting method has a property that human interference is minimized but it has low accuracy. Bootstrap method has a high accuracy, but it has a defect that manual segmentation is required In this paper, a new algorithm is proposed to minimize manual work and to improve the performance of automatic segmentation. At first phase, voiced, unvoiced and silence classification is performed for each speech data frame. At second phase, the phoneme sequence is aligned dynamically to the voiced/unvoiced/silence sequence according to the acoustic phonetic rules. Finally, using these segmented speech data as a bootstrap, phoneme model parameters based on HMM are trained. For the performance test, hand labeled ETRI speech DB was used. The experiment results showed that our algorithm achieved 10% improvement of segmentation accuracy within 20 ms tolerable error range. Especially for the unvoiced consonants, it showed 30% improvement.

• 전자공학회논문지B
• /
• 제33B권9호
• /
• pp.88-96
• /
• 1996

Computer vision-based gesture recognition systems consist of image segmentation, object tracking and decision. However, it is difficult to segment an object from image for gesture in computer systems because of vaious illuminations and backgrounds. In this paper, we describe a method to learn features for segmentation, which improves the performance of computer vision-based hand-gesture recognition systems. Systems interact with a user to acquire exact training data and segment information according to a predefined plan. System provides some models to the user, takes pictures of the user's response and then analyzes the pictures with models and a prior knowledge. The system sends messages to the user and operates learning module to extract information with the analyzed result.

• Journal of International Society for Simulation Surgery
• /
• 제1권1호
• /
• pp.27-31
• /
• 2014

In this paper, an overview of segmentation and 3D visualization methods are presented. Commonly, the two kinds of methods are used to visualize organs and vessels into 3D from medical images such as CT(A) and MRI - Direct Volume Rendering (DVR) and Iso-surface Rendering (IR). DVR can be applied directly to a volume. It directly penetrates through the volume while it determines which voxels are visualizedbased on a transfer function. On the other hand, IR requires a series of processes such as segmentation, polygonization and visualization. To extract a region of interest (ROI) from the medical volume image via the segmentation, some regions of an object and a background are required, which are typically obtained from the user. To visualize the extracted regions, the boundary points of the regions should be polygonized. In other words, the boundary surface composed of polygons such as a triangle and a rectangle should be required to visualize the regions into 3D because illumination effects, which makes the object shaded and seen in 3D, cannot be applied directly to the points.

• 한국멀티미디어학회논문지
• /
• 제23권8호
• /
• pp.965-976
• /
• 2020

We present in this work a segmentation method of E. coli bacterial images generated via phase contrast microscopy using a deep learning based hybrid feature generation. Unlike conventional machine learning methods that use the hand-crafted features, we adopt the denoising autoencoder in order to generate a precise and accurate representation of the pixels. We first construct a hybrid vector that combines original image, difference of Gaussians and image gradients. The created hybrid features are then given to a deep autoencoder that learns the pixels' internal dependencies and the cells' shape and boundary information. The latent representations learned by the autoencoder are used as the inputs of a softmax classification layer and the direct outputs from the classifier represent the coarse segmentation mask. Finally, the classifier's outputs are used as prior information for a graph partitioning based fine segmentation. We demonstrate that the proposed hybrid vector representation manages to preserve the global shape and boundary information of the cells, allowing to retrieve the majority of the cellular patterns without the need of any post-processing.