• The KIPS Transactions:PartB
• /
• v.11B no.3
• /
• pp.387-394
• /
• 2004

In this paper, we propose a modified unsupervised linear alignment algorithm for building an aligned corpus. The original algorithm inserts null characters into both of two aligned strings (source string and target string), because the two strings are different from each other in length. This can cause some difficulties like the search space explosion for applications using the aligned corpus with null characters and no possibility of applying to several machine learning algorithms. To alleviate these difficulties, we modify the algorithm not to contain null characters in the aligned source strings. We have shown the usability of our approach by applying it to different areas such as Korean-English back-trans literation, English grapheme-phoneme conversion, and Korean morphological analysis.

• Journal of KIISE:Software and Applications
• /
• v.35 no.9
• /
• pp.572-587
• /
• 2008

The Self-organizing Feature Map(SOFM) that is one of unsupervised neural networks is a very powerful tool for data clustering and visualization in high-dimensional data sets. Although the SOFM has been applied in many engineering problems, it needs to cluster similar weights into one class on the trained SOFM as a post-processing, which is manually performed in many cases. The traditional clustering algorithms, such as t-means, on the trained SOFM however do not yield satisfactory results, especially when clusters have arbitrary shapes. This paper proposes automatic clustering on trained SOFM, which can deal with arbitrary cluster shapes and be globally optimized by graph cuts. When using the graph cuts, the graph must have two additional vertices, called terminals, and weights between the terminals and vertices of the graph are generally set based on data manually obtained by users. The Proposed method automatically sets the weights based on mode-seeking on a distance matrix. Experimental results demonstrated the effectiveness of the proposed method in texture segmentation. In the experimental results, the proposed method improved precision rates compared with previous traditional clustering algorithm, as the method can deal with arbitrary cluster shapes based on the graph-theoretic clustering.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.13 no.2
• /
• pp.771-789
• /
• 2019

Text detection has been a popular research topic in the field of computer vision. It is difficult for prevalent text detection algorithms to avoid the dependence on datasets. To overcome this problem, we proposed a novel unsupervised text detection algorithm inspired by bootstrap learning. Firstly, the text candidate in a novel form of superpixel is proposed to improve the text recall rate by image segmentation. Secondly, we propose a unique text sample selection model (TSSM) to extract text samples from the current image and eliminate database dependency. Specifically, to improve the precision of samples, we combine maximally stable extremal regions (MSERs) and the saliency map to generate sample reference maps with a double threshold scheme. Finally, a multiple kernel boosting method is developed to generate a strong text classifier by combining multiple single kernel SVMs based on the samples selected from TSSM. Experimental results on standard datasets demonstrate that our text detection method is robust to complex backgrounds and multilingual text and shows stable performance on different standard datasets.

• The Transactions of the Korea Information Processing Society
• /
• v.7 no.8
• /
• pp.2555-2564
• /
• 2000

In this paper, a new unsupervised image segmentation algorithm is proposed. To model the contextual information presented in images, the characteristics of the Markov random fields (MRF) are utilized. Textured images are modeled as realizations of the stationary Gaussian MRF on a two-dimensional square lattice using the conditional autoregressive (CAR) equations with a second-order noncausal neighborhood. To detect boundaries, hypothesis tests over two masked areas are performed. Under the hypothesis, masked areas are assumed to belong to the same class of textures and CAR equation parameters are estimated in a minimum-mean-square-error (MMSE) sense. If the hypothesis is rejected, a measure of dissimilarity between two areas is accumulated on the rejected area. This approach produces potential edge maps. Using these maps, boundary detection can be performed, which resulting no micro edges. The performance of the proposed algorithm is evaluated by some experiments using real images as weB as synthetic ones. The experiments demonstrate that the proposed algorithm can produce satisfactorY segmentation without any a priori information.

• Proceedings of the KSRS Conference
• /
• 2003.11a
• /
• pp.645-647
• /
• 2003

In this work a new method for on-line scene segmentation is developed. In remote sensing a scene is represented by the pixel-oriented features. It is possible to reduce data redundancy by an unsupervised segment-feature extraction process, where the segment-features, rather than the pixelfeatures, are used for multispectral scene representation. The algorithm partitions the observation space into exhaustive set of disjoint segments. Then, pixels belonging to each segment are characterized by segment features. Illustrative examples are presented, and the performance of features is investigated. Results show an average compression more than 25, the classification performance is improved for all classes, and the CPU time required for classification is reduced by the same factor.

• Proceedings of the Korea Institute of Convergence Signal Processing
• /
• 2000.08a
• /
• pp.325-328
• /
• 2000

텍스쳐 분석은 장면 분할, 물체 인식, 모양과 깊이 인식 등의 많은 영상 처리 분야에서 중요한 기술 중의 하나이다. 그러나 실영상에 포함된 다양한 텍스쳐 성분에 대해서 보편적으로 적용 가능한 효율적인 방법들에 대한 연구는 미흡한 실정이다. 본 논문에서는 텍스쳐 인식을 위해서 비교사 학습 방법에 기반 한 효율적인 텍스쳐 분석 기법을 제안한다. 제안된 방법은 텍스쳐 영상이 지닌 방향특징 정보로서 각(angle)과 강도(power)를 추출하여 자기 조직화 신경회로망에 의해서 블록기반으로 군집화(clustering)된다. 비교사적 군집 결과는 통합(merging)과 불림(dilation) 과정을 통해서 영상에 내재된 텍스쳐 성분의 분할을 수행한다. 제안된 시스템의 성능 평가를 위해서는 다양한 형태의 다중 텍스쳐 영상을 생성하여 적용한 후 그 유효성을 보인다.

• Korean Journal of Remote Sensing
• /
• v.22 no.6
• /
• pp.613-619
• /
• 2006

This study proposed an unsupervised image classification through the dendrogram of agglomerative clustering as a higher stage of image segmentation in image processing. The proposed algorithm is a hierarchical clustering which includes searching a set of MCSNP (Mutual Closest Spectral Neighbor Pairs) based on the data structures of RAG(Regional Adjacency Graph) defined on spectral space and Min-Heap. It also employes a multi-window system in spectral space to define the spectral adjacency. RAG is updated for the change due to merging using RNV (Regional Neighbor Vector). The proposed algorithm provides a dendrogram which is a graphical representation of data. The hierarchical relationship in clustering can be easily interpreted in the dendrogram. In this study, the proposed algorithm has been extensively evaluated using simulated images and applied to very large QuickBird imagery acquired over an area of Korean Peninsula. The results have shown it potentiality for the application of remotely-sensed imagery.

• Journal of the Computational Structural Engineering Institute of Korea
• /
• v.36 no.5
• /
• pp.315-321
• /
• 2023

Long-span bridges are flexible structures with low natural frequencies and damping ratios, making them susceptible to vibrational serviceability problems. However, the current design guideline of South Korea assumes a uniform threshold of wind speed or vibrational amplitude to assess the occurrence of harmful vibrations, potentially overlooking the complex vibrational patterns observed in long-span bridges. In this study, we propose a pointwise vortex-induced vibration (VIV) detection method using a deep-learning-based signalsegmentation model. Departing from conventional supervised methods of data acquisition and manual labeling, we synthesize training data by generating sinusoidal waves with an envelope to accurately represent VIV. A Fourier synchrosqueezed transform is leveraged to extract time-frequency features, which serve as input data for training a bidirectional long short-term memory model. The effectiveness of the model trained on synthetic VIV data is demonstrated through a comparison with its counterpart trained on manually labeled real datasets from an actual cable-supported bridge.

• Tunnel and Underground Space
• /
• v.33 no.6
• /
• pp.508-518
• /
• 2023

This paper presents a new approach for the automatic mapping of discontinuities in a tunnel face based on its 3D digital model reconstructed by LiDAR scan or photogrammetry techniques. The main idea revolves around the identification of discontinuity areas in the 3D digital model of a tunnel face by segmenting its 2D projected images using a deep-learning semantic segmentation model called U-Net. The proposed deep learning model integrates various features including the projected RGB image, depth map image, and local surface properties-based images i.e., normal vector and curvature images to effectively segment areas of discontinuity in the images. Subsequently, the segmentation results are projected back onto the 3D model using depth maps and projection matrices to obtain an accurate representation of the location and extent of discontinuities within the 3D space. The performance of the segmentation model is evaluated by comparing the segmented results with their corresponding ground truths, which demonstrates the high accuracy of segmentation results with the intersection-over-union metric of approximately 0.8. Despite still being limited in training data, this method exhibits promising potential to address the limitations of conventional approaches, which only rely on normal vectors and unsupervised machine learning algorithms for grouping points in the 3D model into distinct sets of discontinuities.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.32 no.10C
• /
• pp.990-999
• /
• 2007

In this paper, we propose the region-based image retrieval method using JSEG which is a method for unsupervised segmentation of color-texture regions. JSEG is an algorithm that discretizes an image by color classification, makes the J-image by applying a region to window mask, and then segments the image by using a region growing and merging. The segmented image from JSEG is given to a user as the query image, and a user can select a few segmented regions as the query region. After finding the MBR of regions selected by user query and generating the multiple window masks based on the center point of MBR, we extract the feature vectors from selected regions. We use the accumulated histogram as the global descriptor for performance comparison of extracted feature vectors in each method. Our approach fast and accurately supplies the relevant images for the given query, as the feature vectors extracted from specific regions and global regions are simultaneously applied to image retrieval. Experimental evidence suggests that our algorithm outperforms the recent image-based methods for image indexing and retrieval.