• Journal of Broadcast Engineering
• /
• v.25 no.5
• /
• pp.665-671
• /
• 2020

MPEG-I is actively working on standardization on the coding of immersive video which provides up to 6 degree of freedom (6DoF) in terms of viewpoint. 3DoF+ video, which provides motion parallax to omnidirectional view of 360 video, renders a view at any desired viewpoint using multiple view videos acquisitioned in a limited 3D space covered with upper body motion at a fixed position. The MPEG-I visual group is developing a test model called TMIV (Test Model for Immersive Video) in the process of development of the standard for 3DoF+ video coding. In the TMIV, the redundancy between a set of input view videos is removed, and several atlases are generated by packing patches including the remaining texture and depth regions into frames as compact as possible, and coded. This paper presents an atlas generation method that removes small-sized blocks in the atlas for more efficient 3DoF+ video coding. The proposed method shows a performance improvement of BD-rate bit savings of 0.7% and 1.4%, respectively, in natural and graphic sequences compared to TMIV.

• Journal of Korean Society for Geospatial Information Science
• /
• v.6 no.1 s.11
• /
• pp.11-17
• /
• 1998

Recently, 3D GIS based on 3D geo-processing methodology and Internet environment are emerging issues in GIS fields. To design and implement 3D GIS, the strategic linkage of Java and VRML is first regarded: 3D feature format definition in the passion of conventional GIS including aspatial attributes, 3B feature indexing, 3D analytical operators such as selection, buffering, and Near, Metric operation such as distance measurement and statistical description, and 3D visualization. In 3D feature format definition, the following aspects are implemented: spatial information for 3D primitives extended from 2D primitives, multimedia data, object texture or color of VRML specification. DXF-format GIS layers with additional attributes are converted to 3D feature format and imported into this system. While, 3D analytical operators are realized in the form of 3D buffering with respect to user-defined point, line, polygon, and 3D objects, and 3D Near functions; furthermore, 'Lantern operator' is newly introduced in this 3D GIS. Because this system is implemented by Java applet, any client with Java-enable browser including VRML browser plug-in can utilize the new style of 3D GIS function in the virtual space. Conclusively, we present prototype of WWW-based 3D GIS, and this approach will be contribute to development of core modules on the stage of concept establishment and of real application model in future.

• The KIPS Transactions:PartB
• /
• v.16B no.1
• /
• pp.7-16
• /
• 2009

Some clustering algorithms have a problem that an image is over-segmented since both the spatial information between the segmented regions is not considered and the number of the clusters is defined in advance. Therefore, they are difficult to be applied to the applicable fields. This paper proposes the new post-processing methods, a reclassification of the inhomogeneous clusters and a region merging using Baysian algorithm, that improve the segmentation results of the clustering algorithms. The inhomogeneous cluster is firstly selected based on variance and between-class distance and it is then reclassified into the other clusters in the reclassification step. This reclassification is repeated until the optimal number determined by the minimum average within-class distance. And the similar regions are merged using Baysian algorithm based on Kullbeck-Leibler distance between the adjacent regions. So we can effectively solve the over-segmentation problem and the result can be applied to the applicable fields. Finally, we design a classification system for the segmented regions to validate the proposed method. The segmented regions are classified by SVM(Support Vector Machine) using the principal colors and the texture information of the segmented regions. In experiment, the proposed method showed the validity for various real-images and was effectively applied to the designed classification system.

• Journal of Korean Society of Forest Science
• /
• v.110 no.4
• /
• pp.479-495
• /
• 2021

To improve on the efficient management of forest resources, it is necessary to create a forest soil map, which represents a comprehensive database of forest lands. Although a 1:25,000 scale forest site map has been used in Korea, the need for a large-scale forest soil map with high precision and information on forest lands that is specialized for individual purposes has been identified. Moreover, to keep pace with the advancement in forest management and transition to a digital society, it is essential to develop a method for constructing new forest soil maps that can diversify its use. Therefore, this paper presented a developmental process and used a 1:5,000 scale forest soil map to propose future directions. National maps showing the soil type, depth, and texture were produced based on the survey and analysis of forest soils, followed by the Forest Land Soil Map (1:5,000) Production Standard Manual. Alternatively, forest soil map data were the basis on which various other maps that can be used to prevent and predict forest disasters and evaluate environmental capacities were developed. Accordingly, ways to provide appropriate information to achieve the national forest plan, secure forestry big data, and accomplish sustainable forest management that corresponds to the national development plan are proposed based on results from the current study.

• Journal of the Korea Convergence Society
• /
• v.12 no.11
• /
• pp.45-51
• /
• 2021

With the development of deep learning, semantic segmentation methods are being studied in various fields. There is a problem that segmenation accuracy drops in fields that require accuracy such as medical image analysis. In this paper, we improved PSPNet, which is a deep learning based segmentation method to minimized the loss of features during semantic segmentation. Conventional deep learning based segmentation methods result in lower resolution and loss of object features during feature extraction and compression. Due to these losses, the edge and the internal information of the object are lost, and there is a problem that the accuracy at the time of object segmentation is lowered. To solve these problems, we improved PSPNet, which is a semantic segmentation model. The multi-scale attention proposed to the conventional PSPNet was added to prevent feature loss of objects. The feature purification process was performed by applying the attention method to the conventional PPM module. By suppressing unnecessary feature information, eadg and texture information was improved. The proposed method trained on the Cityscapes dataset and use the segmentation index MIoU for quantitative evaluation. As a result of the experiment, the segmentation accuracy was improved by about 1.5% compared to the conventional PSPNet.

• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.39 no.6
• /
• pp.393-399
• /
• 2021

In this study, experiments were conducted to estimate the PM concentrations by learning the nighttime CCTV images of various PM concentrations environments. In the case of daytime images, there have been many related studies, and the various texture and brightness information of images is well expressed, so the information affecting learning is clear. However, nighttime images contain less information than daytime images, and studies using only nighttime images are rare. Therefore, we conducted an experiment combining nighttime images with non-uniform characteristics due to light sources such as vehicles and streetlights and building roofs, building walls, and streetlights with relatively constant light sources as an ROI (Region of Interest). After that, the correlation was analyzed compared to the daytime experiment to see if deep learning-based PM concentrations estimation was possible with nighttime images. As a result of the experiment, the result of roof ROI learning was the highest, and the combined learning model with the entire image showed more improved results. Overall, R² exceeded 0.9, indicating that PM estimation is possible from nighttime CCTV images, and it was calculated that additional combined learning of weather data did not significantly affect the experimental results.

• Journal of the Korean Association of Geographic Information Studies
• /
• v.27 no.1
• /
• pp.115-127
• /
• 2024

This research aimed to construct models with various structures based on the Transformer module and to perform land cover classification, thereby examining the applicability of the Transformer module. For the classification of land cover, the Unet model, which has a CNN structure, was selected as the base model, and a total of four deep learning models were constructed by combining both the encoder and decoder parts with the Transformer module. During the training process of the deep learning models, the training was repeated 10 times under the same conditions to evaluate the generalization performance. The evaluation of the classification accuracy of the deep learning models showed that the Model D, which utilized the Transformer module in both the encoder and decoder structures, achieved the highest overall accuracy with an average of approximately 89.4% and a Kappa coefficient average of about 73.2%. In terms of training time, models based on CNN were the most efficient. however, the use of Transformer-based models resulted in an average improvement of 0.5% in classification accuracy based on the Kappa coefficient. It is considered necessary to refine the model by considering various variables such as adjusting hyperparameters and image patch sizes during the integration process with CNN models. A common issue identified in all models during the land cover classification process was the difficulty in detecting small-scale objects. To improve this misclassification phenomenon, it is deemed necessary to explore the use of high-resolution input data and integrate multidimensional data that includes terrain and texture information.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.9 no.3
• /
• pp.39-52
• /
• 1999

In this paper we propose a new watermarking technique for copyright protection of images. The proposed technique is based on a spatial masking method with a spatial scale parameter. In general it becomes more robust against various attacks but with some degradations on the image quality as the amplitude of the watermark increases. On the other hand it becomes perceptually more invisible but more vulnerable to various attacks as the amplitude of the watermark decreases. Thus it is quite complex to decide the compromise between the robustness of watermark and its visibility. We note that watermarking using the spread spectrum is not robust enought. That is there may be some areas in the image that are tolerable to strong watermark signals. However large smooth areas may not be strong enough. Thus in order to enhance the invisibility of watermarked image for those areas the spatial masking characteristics of the HVS(Human Visual System) should be exploited. That is for texture regions the magnitude of the watermark can be large whereas for those smooth regions the magnitude of the watermark can be small. As a result the proposed watermarking algorithm is intend to satisfy both the robustness of watermark and the quality of the image. The experimental results show that the proposed algorithm is robust to image deformations(such as compression adding noise image scaling clipping and collusion attack).

• Journal of Korean Society of Forest Science
• /
• v.83 no.2
• /
• pp.175-190
• /
• 1994

In order to get fundamental information for prediction of landslide hazard, both forest and site factors affecting slope stability were investigated in many areas of active landslides. Twelve descriptors were identified and quantified to develop the prediction model by multivariate statistical analysis. The main results obtained could be summarized as follows : The main factors influencing a large scale of landslide were shown in order of precipitation, age group of forest trees, altitude, soil texture, slope gradient, position of slope, vegetation, stream order, vertical slope, bed rock, soil depth and aspect. According to partial correlation coefficient, it was shown in order of age group of forest trees, precipitation, soil texture, bed rock, slope gradient, position of slope, altitude, vertical slope, stream order, vegetation, soil depth and aspect. The main factors influencing a landslide occurrence were shown in order of age group of forest trees, altitude, soil texture, slope gradient, precipitation, vertical slope, stream order, bed rock and soil depth. Two prediction models were developed by magnitude and frequency of landslide. Particularly, a prediction method by magnitude of landslide was changed the score for the convenience of use. If the total store of the various factors mark over 9.1636, it is evaluated as a very dangerous area. The mean score of landslide and non-landslide group was 0.1977 and -0.1977, and variance was 0.1100 and 0.1250, respectively. The boundary value between the two groups related to slope stability was -0.02, and its predicted rate of discrimination was 73%. In the score range of the degree of landslide hazard based on the boundary value of discrimination, class A was 0.3132 over, class B was 0.3132 to -0.1050, class C was -0.1050 to -0.4196, class D was -0.4195 below. The rank of landslide hazard could be divided into classes A, B, C and D by the boundary value. In the number of slope, class A was 68, class B was 115, class C was 65, and class D was 52. The rate of landslide occurrence in class A and class B was shown at the hige prediction of 83%. Therefore, dangerous areas selected by the prediction method of landslide could be mapped for land-use planning and criterion of disaster district. And also, it could be applied to an administration index for disaster prevention.

• The Transactions of the Korea Information Processing Society
• /
• v.2 no.6
• /
• pp.937-946
• /
• 1995

We proposed and efficient block segmentation and classification method for the document analysis using SGLDM(spatial gray level dependence matrix) and BP (back Propagation) neural network. Seven texture features are extracted directly from the SGLDM of each gray-level block image, and by using the nonlinear classifier of neural network BP, we can classify document blocks into 9 categories. The proposed method classifies the equation block, the table block and the flow chart block, which are mostly composed of the characters, out of the blocks that are conventionally classified as non-character blocks. By applying Sobel operator on the gray-level document image beforebinarization, we can reduce the effect of the background noises, and by using the additional horizontal-vertical smoothing as well as the vertical-horizontal smoothing of images, we can obtain an effective block segmentation that does not lead to the segmentation into small pieces. The result of experiment shows that a document can be segmented and classified into the character blocks of large fonts, small fonts, the character recognigible candidates of tables, flow charts, equations, and the non-character blocks of photos, figures, and graphs.