• ETRI Journal
• /
• v.44 no.1
• /
• pp.38-50
• /
• 2022

This paper presents a study on the use of encoder-derived features in decoder-side depth estimation. The scheme of multiview video encoding does not require the transmission of depth maps (which carry the geometry of a three-dimensional scene) as only a set of input views and their parameters are compressed and packed into the bitstream, with a set of features that could make it easier to estimate geometry in the decoder. The paper proposes novel recursive block splitting for the feature extraction process and evaluates different scenarios of feature-driven decoder-side depth estimation, performed by assessing their influence on the bitrate of metadata, quality of the reconstructed video, and time of depth estimation. As efficient encoding of multiview sequences became one of the main scopes of the video encoding community, the experimental results are based on the "geometry absent" profile from the incoming MPEG Immersive video standard. The results show that the quality of synthesized views using the proposed recursive block splitting outperforms that of the state-of-the-art approach.

• Smart Structures and Systems
• /
• v.31 no.4
• /
• pp.365-381
• /
• 2023

The existing vision-based techniques for inspection and condition assessment of civil infrastructure are mostly manual and consequently time-consuming, expensive, subjective, and risky. As a viable alternative, researchers in the past resorted to deep learning-based autonomous damage detection algorithms for expedited post-disaster reconnaissance of structures. Although a number of automatic damage detection algorithms have been proposed, the scarcity of labeled training data remains a major concern. To address this issue, this study proposed a semi-supervised learning (SSL) framework based on consistency regularization and cross-supervision. Image data from post-earthquake reconnaissance, that contains cracks, spalling, and exposed rebars are used to evaluate the proposed solution. Experiments are carried out under different data partition protocols, and it is shown that the proposed SSL method can make use of unlabeled images to enhance the segmentation performance when limited amount of ground truth labels are provided. This study also proposes DeepLab-AASPP and modified versions of U-Net++ based on channel-wise attention mechanism to better segment the components and damage areas from images of reinforced concrete buildings. The channel-wise attention mechanism can effectively improve the performance of the network by dynamically scaling the feature maps so that the networks can focus on more informative feature maps in the concatenation layer. The proposed DeepLab-AASPP achieves the best performance on component segmentation and damage state segmentation tasks with mIoU scores of 0.9850 and 0.7032, respectively. For crack, spalling, and rebar segmentation tasks, modified U-Net++ obtains the best performance with Igou scores (excluding the background pixels) of 0.5449, 0.9375, and 0.5018, respectively. The proposed architectures win the second place in IC-SHM2021 competition in all five tasks of Project 2.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.22 no.5
• /
• pp.185-191
• /
• 2022

In this study, we propose a novel super-resolution neural network based on U-Net, residual neural network, and sub-pixel convolution. To prevent the loss of detailed information due to the max pooling of U-Net, we propose down-sampling and connection using sub-pixel convolution. This uses all pixels in the filter, unlike the max pooling that creates a new feature map with only the max value in the filter. As a 2×2 size filter passes, it creates a feature map consisting only of pixels in the upper left, upper right, lower left, and lower right. This makes it half the size and quadruple the number of feature maps. And we propose two methods to reduce the computation. The first uses sub-pixel convolution, which has no computation, and has better performance, instead of up-convolution. The second uses a layer that adds two feature maps instead of the connection layer of the U-Net. Experiments with a banchmark dataset show better PSNR values on all scale and benchmark datasets except for set5 data on scale 2, and well represent local area patterns.

• Journal of Broadcast Engineering
• /
• v.25 no.3
• /
• pp.362-373
• /
• 2020

Recently, as autonomous vehicles and intelligent CCTV are growing more interest, the efficient object detection is essential technique. The DPM(Deformable Part Models) which is basis of this paper have used a typical object system that represents highly variable objects using mixtures of deformable part for object. Although it shows high detection performance by capturing part shape and configuration of object model, but it is limited to use in real application due to the complicated algorithm. In this paper, instead of image feature pyramid that takes up a large amount of computation in one part of the detector, we propose a method to reduce the computation speed by reconstructing a new image feature pyramid that uses adaptive bilinear interpolation of feature maps obtained on a specific image scale. As a result, the detection performance for object was lowered a little by 2.82%, however, the proposed detection method improved the speed performance by 10% in comparison with original DPM.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.9 no.3
• /
• pp.1121-1139
• /
• 2015

In texture-plus-depth format, depth map compression is an important task. Different from normal texture images, depth maps have less texture information, while contain many homogeneous regions separated by sharp edges. This feature will be employed to form an efficient depth map coding scheme in this paper. Firstly, the histogram of the depth map will be analyzed to find an appropriate threshold that segments the depth map into the foreground and background regions, allowing the edge between these two kinds of regions to be obtained. Secondly, the two regions will be encoded through rate distortion optimization with a shape adaptive wavelet transform, while the edges are lossless encoded with JBIG2. Finally, a depth-updating algorithm based on the threshold and the depth range is applied to enhance the quality of the decoded depth maps. Experimental results demonstrate the effective performance on both the depth map quality and the synthesized view quality.

• Proceedings of the KSRS Conference
• /
• 1999.11a
• /
• pp.386-389
• /
• 1999

In this study, visible and near infrared channels of NOAA/AVHRR data were used to classify land use and vegetation types over Korean peninsula. Analyzing forest stand structures and prediction of forest productivity using satellite data were also reviewed. Land use and land cover classification was made by unsupervised clustering methods. After monthly Normalized Difference Vegetation Index (NDVI) composite images were derived from April to November 1998, the derived composite images were used as temporal feature vector's in this clustering analysis. Visually interpreted, the classification result was satisfactory in overall for it matched well with the general land cover patterns. But subclassification of forests into coniferous, deciduous, and mixed forests were much confused due to the effects of low ground resolution of AVHRR data and without defined classification scheme. To investigate into the forest stand structures, digital forest type maps were used as an ancillary data. Forest type maps, which were compiled and digitalized by Forestry Research Institute, were registered to AVHRR image coordinates. Two data sets were compared and percent forest cover over whole region was estimated by multiple regression analysis. Using this method, other forest stand structure characteristics within the primary data pixels are expected to be extracted and estimated.

• Journal of the Korea Institute of Military Science and Technology
• /
• v.12 no.2
• /
• pp.199-205
• /
• 2009

A simple and effective analysis method is presented for predicting the best infiltration area. Based on geospatial information, numerical estimation barometer for degree of infiltration risk has been derived. The dominant geospatial features influencing infiltration risk have been found to be area altitude, degree of surface gradient, relative direction of surface gradient to the surveillance line, degree of surface gradient repetition, regional forest information. Each feature has been numerically expressed corresponding to the degree of infiltration risk of that area. Four different detection probability maps of infiltration risk for the surveillance area are drawn on the actual map with respect to the numerically expressed five dominant factors of infiltration risks. By combining the four detection probability maps, the complete picture of thr best infiltration area has been drawn. By using the map and the analytic method the effectiveness of surveillance operation can be improved.

• Bulletin of the Korean Chemical Society
• /
• v.28 no.3
• /
• pp.379-385
• /
• 2007

Receptor-oriented pharmacophore-based in silico screening is a powerful tool for rapidly screening large number of compounds for interactions with a given protein. Inhibition of the enzyme catechol-Omethyltransferase (COMT) offers a novel possibility for treating Parkinson's disease. Bisubstrate inhibitors of COMT containing the adenine of S-adenosylmethionine (SAM) and a catechol moiety are a new class of potent and selective inhibitor. In the present study, we used receptor-oriented pharmacophore-based in silico screening to examine the interactions between the active site of human COMT and bisubstrate inhibitors. We generated 20 pharmacophore maps, of which 4 maps reproduced the docking model of hCOMT and a bisubstrate inhibitor. Only one of these four, pharmacophore map I, effectively described the common features of a series of bisubstrate inhibitors. Pharmacophore map I consisted of one hydrogen bond acceptor (to Mg2+), three hydrogen bond donors (to Glu199, Glu90, and Gln120), and one hydrophobic feature (an active site region surrounded by several aromatic and hydrophobic residues). This map represented the most essential pharmacophore for explaining interactions between hCOMT and a bisubstrate inhibitor. These results revealed a pharmacophore that should help in the development of new drugs for treating Parkinson's disease.

• The KIPS Transactions:PartD
• /
• v.8D no.5
• /
• pp.507-512
• /
• 2001

Feature-based similarity has become an important research issue in multimedia database systems. The features of multimedia data are useful for discriminating between multimedia objects. the performance of conventional multidimensional data structures tends to deteriorate as the number of dimensions of feature vectors increase. The $R^{*}-Tree$ is the most successful variant of the R-Tree. In this paper, we propose a SOM-based $R^{*}-Tree$ as a new indexing method for high-dimensional feature vectors. The SOM-based $R^{*}-Tree$ combines SOM and $R^{*}-Tree$ to achieve search performance more scalable to high-dimensionalties. Self-Organizingf Maps (SOMs) provide mapping from high-dimensional feature vectors onto a two-dimensional space. The map is called a topological feature map, and preserves the mutual relationships (similarity) in the feature spaces of input data, clustering mutually similar feature vectors in neighboring nodes. Each node of the topological feature map holds a codebook vector. We experimentally compare the retrieval time cost of a SOM-based $R^{*}-Tree$ with of an SOM and $R^{*}-Tree$ using color feature vectors extracted from 40,000 images. The results show that the SOM-based $R^{*}-Tree$ outperform both the SOM and $R^{*}-Tree$ due to reduction of the number of nodes to build $R^{*}-Tree$ and retrieval time cost.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.14 no.10
• /
• pp.3955-3971
• /
• 2020

Due to the significant increase in the use of autonomous car technology, it is essential to integrate this technology with high-precision digital map data containing more precise and accurate roadway information, as compared to existing conventional map resources, to ensure the safety of self-driving operations. While existing map technologies may assist vehicles in identifying their locations via Global Positioning System, it is however difficult to update the environmental changes of roadways in these maps. Roadway vision algorithms can be useful for building autonomous vehicles that can avoid accidents and detect real-time location changes. We incorporate a hybrid architectural design that combines unsupervised classification of vision data with supervised joint fusion classification to achieve a better noise-resistant algorithm. We identify, via a deep learning approach, an intelligent hybrid fusion algorithm for fusing multimodal vision feature data for roadway classifications and characterize its improvement in accuracy over unsupervised identifications using image processing and supervised vision classifiers. We analyzed over 93,000 vision frame data collected from a test vehicle in real roadways. The performance indicators of the proposed hybrid fusion algorithm are successfully evaluated for the generation of roadway digital maps for autonomous vehicles, with a recall of 0.94, precision of 0.96, and accuracy of 0.92.