• The Journal of the Acoustical Society of Korea
• /
• v.18 no.6
• /
• pp.49-56
• /
• 1999

In many continuous speech recognition systems based on HMMs, decision tree-based state tying has been used for not only improving the robustness and accuracy of context dependent acoustic modeling but also synthesizing unseen models. To construct the phonetic decision tree, standard method performs one-level pruning using just single Gaussian triphone models. In this paper, two novel approaches, two-level decision tree and multi-mixture decision tree, are proposed to get better performance through more accurate acoustic modeling. Two-level decision tree performs two level pruning for the state tying and the mixture weight tying. Using the second level, the tied states can have different mixture weights based on the similarities in their phonetic contexts. In the second approach, phonetic decision tree continues to be updated with training sequence, mixture splitting and re-estimation. Multi-mixture Gaussian as well as single Gaussian models are used to construct the multi-mixture decision tree. Continuous speech recognition experiment using these approaches on BN-96 and WSJ5k data showed a reduction in word error rate comparing to the standard decision tree based system given similar number of tied states.

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.44 no.4 s.316
• /
• pp.28-35
• /
• 2007

Airbone laser altimeters have been utilized for 3D topographic mapping of the earth, moon, and planets with high resolution and accuracy, which is a rapidly growing remote sensing technique that measures the round-trip time emitted laser pulse to determine the topography. The traveling time from the laser scanner to the Earth's surface and back is directly related to the distance of the sensor to the ground. When there are several objects within the travel path of the laser pulse, the reflected laser pluses are distorted by surface variation within the footprint, generating multiple echoes because each target transforms the emitted pulse. The shapes of the received waveforms also contain important information about surface roughness, slope and reflectivity. Waveform processing algorithms parameterize and model the return signal resulting from the interaction of the transmitted laser pulse with the surface. Each of the multiple targets within the footprint can be identified. Assuming each response is gaussian, returns are modeled as a mixture gaussian distribution. Then, the parameters of the model are estimated by LMS Method or EM algorithm However, each response actually shows the skewness in the right side with the slowly decaying tail. For the application to require more accurate analysis, the tail information is to be quantified by an approach to decompose the tail. One method to handle with this problem is proposed in this study.

• Journal of Korea Multimedia Society
• /
• v.15 no.9
• /
• pp.1075-1085
• /
• 2012

In clinical research using medical images, the image segmentation is one of the most important processes. Especially, the hippocampal atrophy is helpful for the clinical Alzheimer diagnosis as a specific marker of the progress of Alzheimer. In order to measure hippocampus volume exactly, segmentation of the hippocampus is essential. However, the hippocampus has some features like relatively low contrast, low signal-to-noise ratio, discreted boundary in MRI images, and these features make it difficult to segment hippocampus. To solve this problem, firstly, We selected region of interest from an experiment image, subtracted a original image from the negative image of the original image, enhanced contrast, and applied anisotropic diffusion filtering and gaussian filtering as preprocessing. Finally, We performed an image segmentation using two level set methods. Through a variety of approaches for the validation of proposed hippocampus segmentation method, We confirmed that our proposed method improved the rate and accuracy of the segmentation. Consequently, the proposed method is suitable for segmentation of the area which has similar features with the hippocampus. We believe that our method has great potential if successfully combined with other research findings.

• The Journal of Korea Institute of Information, Electronics, and Communication Technology
• /
• v.12 no.3
• /
• pp.321-326
• /
• 2019

Image denoising methods based on block matching are founded on the experimental observations that neighboring patches or blocks in images retain similar features with each other, and have been proved to show superior performance in denoising different kinds of noise. The methods, however, take into account only neighboring blocks in searching for similar blocks, and ignore the characteristic features of the reference block itself. Consequently, denoising performance is negatively affected when outliers of the Gaussian distribution are included in the reference block which is to be denoised. In this paper, we propose an expanded block matching method in which noisy images are first decomposed into a number of bit-planes, then the range of true signals are estimated based on the distribution of pixels on the bit-planes, and finally outliers are replaced by the neighboring pixels belonging to the estimated range. In this way, the advantages of the conventional Gaussian filter can be added to the blocking matching method. We tested the proposed method through extensive experiments with well known test-bed images, and observed that performance gain can be achieved by the proposed method.

• The KIPS Transactions:PartA
• /
• v.13A no.5 s.102
• /
• pp.465-472
• /
• 2006

Recently, many suggestions have been made in image segmentation methods for extracting human organs or disease affected area from huge amounts of medical image datasets. However, images from some areas, such as brain, which have multiple structures with ambiruous structural borders, have limitations in their structural segmentation. To address this problem, clustering technique which classifies voxels into finite number of clusters is often employed. This, however, has its drawback, the influence from noise, which is caused from voxel by voxel operations. Therefore, applying image enhancing method to minimize the influence from noise and to make clearer image borders would allow more robust structural segmentation. This research proposes an efficient structural segmentation method by filtering based clustering to extract detail structures such as white matter, gray matter and cerebrospinal fluid from brain MR. First, coherence enhancing diffusion filtering is adopted to make clearer borders between structures and to reduce the noises in them. To the enhanced images from this process, fuzzy c-means clustering method was applied, conducting structural segmentation by assigning corresponding cluster index to the structure containing each voxel. The suggested structural segmentation method, in comparison with existing ones with clustering using Gaussian or general anisotropic diffusion filtering, showed enhanced accuracy which was determined by how much it agreed with the manual segmentation results. Moreover, by suggesting fine segmentation method on the border area with reproducible results and minimized manual task, it provides efficient diagnostic support for morphological abnormalities in brain.

• The KIPS Transactions:PartB
• /
• v.12B no.3 s.99
• /
• pp.239-246
• /
• 2005

Biologist must have to do 2DGE biological experiment for Protein Search and Analysis. This experiment coming into being 2 dimensional image. 2DGE (2D Gel Electrophoresis : two dimensional gel electrophoresis) image is the most widely used method for isolating of the objective protein by comparative analysis of the protein spot pattern in the gel plane. The process of protein spot analysis, firstly segment protein spots that are spread in 2D gel plane by image processing and can find important protein spots through comparative analysis with protein pattern of contrast group. In the algorithm which detect protein spots, previous 2DGE image analysis is applies gaussian fitting, however recently Watersheds region based segmentation algorithm, which is based on morphological segmentation is applied. Watersheds has the benefit that segment rapidly needed field in big sized image, however has under-segmentation and over-segmentation of spot area when gray level is continuous. The drawback was somewhat solved by marker point institution, but needs the split and merge process. This paper introduces a novel marker search of protein spots by watersheds-based hierarchical threshold, which can resolve the problem of marker-driven watersheds.

• Journal of the Korean Institute of Telematics and Electronics T
• /
• v.36T no.3
• /
• pp.26-31
• /
• 1999

In this paper, we propose a facial characteristic detection algorithm based on knowledge and object segmentation method for image communication. In this algorithm, under the condition of the same lumination and distance from the fixed video camera to human face, we capture input images of 256 $\times$ 256 of gray scale 256 level and then remove the noise using the Gaussian filter. Two images are captured with a video camera, One contains the human face; the other contains only background region without including a face. And then we get a differential image between two images. After removing noise of the differential image by eroding End dilating, divide background image into a facial image. We separate eyes, ears, a nose and a mouth after searching the edge component in the facial image. From simulation results, we have verified the efficiency of the Proposed algorithm.

• Journal of Digital Convergence
• /
• v.10 no.4
• /
• pp.191-197
• /
• 2012

In this paper we present creating 3D image from 2D image by extract initial depth values calculated from focal values. The initial depth values are created by using the extracted focal information, which is calculated by the comparison of original image and Gaussian filtered image. This initial depth information is allocated to the object segments obtained from normalized cut technique. Then the depth of the objects are corrected to the average of depth values in the objects so that the single object can have the same depth. The generated depth is used to convert to 3D image using DIBR(Depth Image Based Rendering) and the generated 3D image is compared to the images generated by other techniques.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.39C no.6
• /
• pp.490-496
• /
• 2014

The thickness of knee joint cartilage causes most diseases of knee. Therefore, an articular cartilage segmentation of knee magnetic resonance imaging (MRI) is required to diagnose a knee diagnosis correctly. In particular, fully automatic segmentation method of knee joint cartilage enables an effective diagnosis of knee disease. In this paper, we analyze a well-known level-set based segmentation method in brain MRI, and apply that method to knee MRI with solving some problems from different image characteristics. The proposed method, a fully automatic segmentation in whole process, enables to process faster than previous semi-automatic segmentation methods. Also it can make a three-dimension visualization which provides a specialist with an assistance for the diagnosis of knee disease. In addition, the proposed method provides more accurate results than the existing methods of articular cartilage segmentation in knee MRI through experiments.

• The KIPS Transactions:PartB
• /
• v.18B no.3
• /
• pp.131-138
• /
• 2011

In this paper, we compare our proposed method with previous methods for the volumetric image segmentation using level set. In order to obtain an exact segmentation, the region and boundary information of image object are used in our proposed speed function. The boundary information is defined by the gradient vector flow obtained from the gradient images and the region information is defined by Gaussian distribution information of pixel intensity in a region-of-interest for image segmentation. Also the regular term is used to remove the noise around surface. We show various experimental results of real medical volume images to verify the superiority of proposed method.