• Journal of Broadcast Engineering
• /
• v.27 no.1
• /
• pp.69-79
• /
• 2022

In this study, a method of generating and utilizing a normal map image used to represent the characteristics of the surface of an image material to improve the classification accuracy of the original material image is proposed. First of all, (1) to generate a normal map that reflects the surface properties of a material in an image, a U-Net with attention-R2 gate as a generator was used, and a Pix2Pix-based method using the generated normal map and the similarity with the original normal map as a reconstruction loss was used. Next, (2) we propose a network that can improve the accuracy of classification of the original material image by applying the previously created normal map image to the attention gate of the classification network. For normal maps generated using Pixar Dataset, the similarity between normal maps corresponding to ground truth is evaluated. In this case, the results of reconstruction loss function applied differently according to the similarity metrics are compared. In addition, for evaluation of material image classification, it was confirmed that the proposed method based on MINC-2500 and FMD datasets and comparative experiments in previous studies could be more accurately distinguished. The method proposed in this paper is expected to be the basis for various image processing and network construction that can identify substances within an image.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.17 no.2
• /
• pp.15-23
• /
• 2017

The person Re-identification is the most challenging part of computer vision due to the significant changes in human pose and background clutter with occlusions. The picture from non-overlapping cameras enhance the difficulty to distinguish some person from the other. To reach a better performance match, most methods use feature selection and distance metrics separately to get discriminative representations and proper distance to describe the similarity between person and kind of ignoring some significant features. This situation has encouraged us to consider a novel method to deal with this problem. In this paper, we proposed an enhanced recurrent neural network with three-tier hierarchical network for person re-identification. Specifically, the proposed recurrent neural network (RNN) model contain an iterative expectation maximum (EM) algorithm and three-tier Hierarchical network to jointly learn both the discriminative features and metrics distance. The iterative EM algorithm can fully use of the feature extraction ability of convolutional neural network (CNN) which is in series before the RNN. By unsupervised learning, the EM framework can change the labels of the patches and train larger datasets. Through the three-tier hierarchical network, the convolutional neural network, recurrent network and pooling layer can jointly be a feature extractor to better train the network. The experimental result shows that comparing with other researchers' approaches in this field, this method also can get a competitive accuracy. The influence of different component of this method will be analyzed and evaluated in the future research.

• Journal of Navigation and Port Research
• /
• v.47 no.6
• /
• pp.367-375
• /
• 2023

This study developed a new distance metric for vessel trajectories, applicable to marine traffic control services in the Korean coastal waters. The proposed metric is designed through the weighted summation of the traditional Hausdorff distance, which measures the similarity between spatiotemporal data and incorporates the differences in the average Speed Over Ground (SOG) and the variance in Course Over Ground (COG) between two trajectories. To validate the effectiveness of this new metric, a comparative analysis was conducted using the actual Automatic Identification System (AIS) trajectory data, in conjunction with an agglomerative clustering algorithm. Data visualizations were used to confirm that the results of trajectory clustering, with the new metric, reflect geographical distances and the distribution of vessel behavioral characteristics more accurately, than conventional metrics such as the Hausdorff distance and Dynamic Time Warping distance. Quantitatively, based on the Davies-Bouldin index, the clustering results were found to be superior or comparable and demonstrated exceptional efficiency in computational distance calculation.

• Transactions of the Korean Society of Mechanical Engineers A
• /
• v.40 no.4
• /
• pp.397-406
• /
• 2016

Diverse stakeholders engaged in design, construction, and operation and maintenance of offshore plants typically operate heterogeneous plant 3D CAD systems. Engineering, procurement, and construction (EPC) companies are required to submit plant design result to the owner in the form of a plant 3D CAD model, as specified in the contract. However, because of the limitations of data interface of plant 3D CAD systems, EPC companies frequently perform manual remodeling to fulfill the terms and conditions of the contract. Therefore, comparison should be performed between the source plant 3D CAD model and the remodeled plant 3D CAD model to prove the validity of the remodeled plant 3D CAD model. To automate the comparison process, we have developed a system for quantitatively assessing the similarity of the plant 3D CAD models. This paper presents the architecture and detailed functions of the system. In addition, experimental results using this system are explained.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.14 no.5
• /
• pp.2122-2140
• /
• 2020

Proper and precise embedding of commercial ads within Webpages requires Ad-hoc analysis and understanding of their content. By the successful implementation of this step, both publishers and advertisers gain mutual benefits through increasing their revenues on the one hand, and improving user experience on the other. In this research work, we propose a novel multi-level context-based ads serving approach through which ads will be served at generic publisher websites based on their contextual relevance. In the proposed approach, knowledge encoded in domain-specific and generic semantic repositories is exploited in order to analyze and segment Webpages into sets of contextually-relevant segments. Semantically-enhanced indexes are also constructed to index ads based on their textual descriptions provided by advertisers. A modified cosine similarity matching algorithm is employed to embed each ad from the Ads repository into one or more contextually-relevant segments. In order to validate our proposal, we have implemented a prototype of an ad serving system with two datasets that consist of (11429 ads and 93 documents) and (11000 documents and 15 ads), respectively. To demonstrate the effectiveness of the proposed techniques, we experimentally tested the proposed method and compared the produced results against five baseline metrics that can be used in the context of ad serving systems. In addition, we compared the results produced by our system with other state-of-the-art models. Findings demonstrate that the accuracy of conventional ad matching techniques has improved by exploiting the proposed semantically-enhanced context-based ad serving model.

• Journal of the Korea Society of Computer and Information
• /
• v.15 no.11
• /
• pp.67-73
• /
• 2010

In this paper, a new face detection method that is more accurate than the conventional methods is proposed. This method utilizes a hybrid of Hausdorff distance based on the geometric similarity between the two sets of points and the LBP distance based on the distribution of local micro texture of an image. The parameters for normalization and the optimal blending factor of the two different metrics were calculated from training sample images. Popularly used face database was used to show that the proposed method is more effective and robust to the variation of the pose, illumination, and back ground than the methods based on the Hausdorff distance or LBP distance. In the particular case, the average error distance between the detected and the true face location was reduced to 47.9% of the result of LBP method, and 22.8% of the result of Hausdorff method.

• Investigative Magnetic Resonance Imaging
• /
• v.24 no.3
• /
• pp.123-131
• /
• 2020

Purpose: Joint label fusion (JLF) is a popular multi-atlas-based segmentation algorithm, which compensates for dependent errors that may exist between atlases. However, in order to get good segmentation results, it is very important to set the several free parameters of the algorithm to optimal values. In this study, we first investigate the feasibility of a JLF algorithm for prostate segmentation in MR images, and then suggest the optimal set of parameters for the automatic prostate segmentation by validating the results of each parameter combination. Materials and Methods: We acquired T2-weighted prostate MR images from 20 normal heathy volunteers and did a series of cross validations for every set of parameters of JLF. In each case, the atlases were rigidly registered for the target image. Then, we calculated their voting weights for label fusion from each combination of JLF's parameters (r_pxy, r_pz, r_sxy, r_sz, β). We evaluated the segmentation performances by five validation metrics of the Prostate MR Image Segmentation challenge. Results: As the number of voxels participating in the voting weight calculation and the number of referenced atlases is increased, the overall segmentation performance is gradually improved. The JLF algorithm showed the best results for dice similarity coefficient, 0.8495 ± 0.0392; relative volume difference, 15.2353 ± 17.2350; absolute relative volume difference, 18.8710 ± 13.1546; 95% Hausdorff distance, 7.2366 ± 1.8502; and average boundary distance, 2.2107 ± 0.4972; in parameters of r_pxy = 10, r_pz = 1, r_sxy = 3, r_sz = 1, and β = 3. Conclusion: The evaluated results showed the feasibility of the JLF algorithm for automatic segmentation of prostate MRI. This empirical analysis of segmentation results by label fusion allows for the appropriate setting of parameters.

• Korean Journal of Remote Sensing
• /
• v.31 no.5
• /
• pp.433-440
• /
• 2015

In stochastic hyperspectral target detection algorithms, the target signal components may be included in the background characterization if targets are not rare in the image, causing target leakage. In this paper, the effect of target leakage is analysed and an improved hyperspectral target detection method is proposed by excluding the pixels which have similar reflectance spectrum with the target in the process of background characterization. Experimental results using the AISA airborne hyperspectral data and simulated data with artificial targets show that the proposed method can dramatically improve the target detection performance of matched filter and adaptive cosine estimator. More studies on the various metrics for measuring spectral similarity and adaptive method to decide the appropriate amount of exclusion are expected to increase the performance and usability of this method.

• Korean Journal of Materials Research
• /
• v.32 no.8
• /
• pp.345-353
• /
• 2022

In this study, using deep learning, super-resolution images of transmission electron microscope (TEM) images were generated for nanomaterial analysis. 1169 paired images with 256 × 256 pixels (high resolution: HR) from TEM measurements and 32 × 32 pixels (low resolution: LR) produced using the python module openCV were trained with deep learning models. The TEM images were related to DyVO₄ nanomaterials synthesized by hydrothermal methods. Mean-absolute-error (MAE), peak-signal-to-noise-ratio (PSNR), and structural similarity (SSIM) were used as metrics to evaluate the performance of the models. First, a super-resolution image (SR) was obtained using the traditional interpolation method used in computer vision. In the SR image at low magnification, the shape of the nanomaterial improved. However, the SR images at medium and high magnification failed to show the characteristics of the lattice of the nanomaterials. Second, to obtain a SR image, the deep learning model includes a residual network which reduces the loss of spatial information in the convolutional process of obtaining a feature map. In the process of optimizing the deep learning model, it was confirmed that the performance of the model improved as the number of data increased. In addition, by optimizing the deep learning model using the loss function, including MAE and SSIM at the same time, improved results of the nanomaterial lattice in SR images were achieved at medium and high magnifications. The final proposed deep learning model used four residual blocks to obtain the characteristic map of the low-resolution image, and the super-resolution image was completed using Upsampling2D and the residual block three times.

• ETRI Journal
• /
• v.44 no.4
• /
• pp.573-587
• /
• 2022

The agricultural sector is completely different from other sectors since it completely relies on various natural and climatic factors. Climate changes have many effects, including lack of annual rainfall and pests, heat waves, changes in sea level, and global ozone/atmospheric CO₂ fluctuation, on land and agriculture in similar ways. Climate change also affects the environment. Based on these factors, farmers chose their crops to increase productivity in their fields. Many existing agricultural ontologies are either domain-specific or have been created with minimal vocabulary and no proper evaluation framework has been implemented. A new agricultural ontology focused on subdomains is designed to assist farmers using Jaccard relative extractor (JRE) and Naïve Bayes algorithm. The JRE is used to find the similarity between two sentences and words in the agricultural documents and the relationship between two terms is identified via the Naïve Bayes algorithm. In the proposed method, the preprocessing of data is carried out through natural language processing techniques and the tags whose dimensions are reduced are subjected to rule-based formal concept analysis and mapping. The subdomain ontologies of weather, pest, and soil are built separately, and the overall agricultural ontology are built around them. The gold standard for the lexical layer is used to evaluate the proposed technique, and its performance is analyzed by comparing it with different state-of-the-art systems. Precision, recall, F-measure, Matthews correlation coefficient, receiver operating characteristic curve area, and precision-recall curve area are the performance metrics used to analyze the performance. The proposed methodology gives a precision score of 94.40% when compared with the decision tree(83.94%) and K-nearest neighbor algorithm(86.89%) for agricultural ontology construction.