• Journal of the Korea Society of Computer and Information
• /
• v.27 no.2
• /
• pp.33-42
• /
• 2022

In this paper, we proposed a deep learning model based on multi-scale dilated convolution feature fusion for the segmentation of black ice region in road image to send black ice warning to drivers in real time. In the proposed multi-scale dilated convolution feature fusion network, different dilated ratio convolutions are connected in parallel in the encoder blocks, and different dilated ratios are used in different resolution feature maps, and multi-layer feature information are fused together. The multi-scale dilated convolution feature fusion improves the performance by diversifying and expending the receptive field of the network and by preserving detailed space information and enhancing the effectiveness of diated convolutions. The performance of the proposed network model was gradually improved with the increase of the number of dilated convolution branch. The mIoU value of the proposed method is 96.46%, which was higher than the existing networks such as U-Net, FCN, PSPNet, ENet, LinkNet. The parameter was 1,858K, which was 6 times smaller than the existing LinkNet model. From the experimental results of Jetson Nano, the FPS of the proposed method was 3.63, which can realize segmentation of black ice field in real time.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.26 no.1
• /
• pp.44-49
• /
• 2016

This paper proposed the method to improve performance of a pixel, which has low accuracy, by applying image segmentation methods based on color and multi-threshold of brightness. Stereo matching is the process to find the corresponding point on the right image with the point on the left image. For this process, distance(depth) information in stereo images is calculated. However, in the case of a region which has textureless, stereo matching has low accuracy and bad pixels occur on the disparity map. In the proposed method, the relationship between adjacent pixels is considered for compensating bad pixels. Generally, the object has similar color and brightness. Therefore, by considering the relationship between regions based on segmented regions by means of color and multi-threshold of brightness respectively, the region which is considered as parts of same object is re-segmented. According to relationship information of segmented sets of pixels, bad pixels in the disparity map are compensated efficiently. By applying the proposed method, the results show a decrease of nearly 28% in the number of bad pixels of the image applied the method which is established.

• Journal of Internet Computing and Services
• /
• v.13 no.1
• /
• pp.125-134
• /
• 2012

As the social networking service (SNS), Facebook, and Cyworld, is developing, the social network game and social business commerce based on this service is activated. Especially, the Social Network Game (SNG) is getting explosive interests and it becomes popular, because it is small scale and user can enjoy the game among close friends. The market for this game is getting larger every year, but still it has some limitations in developing the game. Especially, the current game engine is aiming for developing online or console game, and there is no exclusive game engine for developing SNG. Therefore, it takes lots of time for developing SNG with this game engine. In this paper, we described a design and development of the game engine optimized for developing SNG, which not only adapts the main characteristics of the previous game engine, but also considers the specific characteristics of the SNG. The engine also supports map for the simulation game that is the most popular game in SNG, and also provides modules and tools for developing character animation easily. The evaluation standard for the performance of the game engine is the output generation speed of image, text and character. And the results showed reasonable output speed for developing the SNG in generation of image, text, and character.

• Journal of Broadcast Engineering
• /
• v.26 no.4
• /
• pp.429-440
• /
• 2021

With the recent development of deep convolutional neural network learning, deep learning techniques applied to single image super-resolution are showing good results. One of the existing deep learning-based super-resolution techniques is RDN(Residual Dense Network), in which the initial feature information is transmitted to the last layer using residual dense blocks, and subsequent layers are restored using input information of previous layers. However, if all hierarchical features are connected and learned and a large number of residual dense blocks are stacked, despite good performance, a large number of parameters and huge computational load are needed, so it takes a lot of time to learn a network and a slow processing speed, and it is not applicable to a mobile system. In this paper, we use the residual dense structure, which is a continuous memory structure that reuses previous information, and the residual dense channel attention block using the channel attention method that determines the importance according to the feature map of the image. We propose a method that can increase the depth to obtain a large receptive field and maintain a concise model at the same time. As a result of the experiment, the proposed network obtained PSNR as low as 0.205dB on average at 4× magnification compared to RDN, but about 1.8 times faster processing speed, about 10 times less number of parameters and about 1.74 times less computation.

• Journal of Broadcast Engineering
• /
• v.16 no.4
• /
• pp.573-583
• /
• 2011

In this paper, we propose an occlusion compensation algorithm which is used for virtual view generation. In general, since occlusion region is recovered from neighboring pixels by taking the mean value or median value of neighbor pixels, the visual characteristics of a given image are not considered and consequently the accuracy of the compensated occlusion regions is not guaranteed. To solve these problem, we propose an algorithm that considers primary visual characteristics of a given image to compensate the occluded regions by using seam carving algorithm. In the proposed algorithm, we first use Sobel mask to obtain the edge map of a given image and then make it binary digit 0 or 1 and finally thinning process follows. Then, the energy patterns of original and thinned edge map obtained by the modified seam carving method are used to compensate the occlusion regions. Through experiments with many test images, we verify that the proposed algorithm performed better than conventional algorithms.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.20 no.3
• /
• pp.153-161
• /
• 2020

Due to the recent development of the 4th industrial revolution, Smart Factories that organically link various technologies such as AI, IoT, Cloud, and Big Data are increasing. Based on this, in the industrial environment where the internal process is controlled automatically, high availability should be secured against the loss caused when the internal process of the Smart Factory is stopped due to the determinism and malicious attack necessary to control the device such as PLC. The research and analysis of industrial network equipment and security equipment used in various industries can improve the efficiency and usability of industrial control systems in national infrastructure and can provide important feedback to build related infrastructure. Therefore, we compared industrial network equipment and security equipment in this paper in a variety of ways and expect to be used as a roadmap for developing technologies for industrial network equipment and industrial security equipment based on the results of this paper.

• The Journal of the Korea Contents Association
• /
• v.12 no.12
• /
• pp.63-70
• /
• 2012

In medical image processing field, object recognition is usually processed based on region segmentation algorithm. Region segmentation in the computing field is carried out by computerized processing of various input information such as brightness, shape, and pattern analysis. If the information mentioned does not make sense, however, many limitations could occur with region segmentation during computer processing. Therefore, this paper suggests effective region segmentation method based on R2-map information within the magnetic resonance (MR) theory. In this study, the experiment had been conducted using images including the liver region and by setting up feature points of R2-map as seed points for 2D region growing and final boundary correction to enable region segmentation even when the border line was not clear. As a result, an average area difference of 7.5%, which was higher than the accuracy of conventional exist region segmentation algorithm, was obtained.

• Journal of the Korea Society of Computer and Information
• /
• v.27 no.3
• /
• pp.251-258
• /
• 2022

In this paper, we propose a standard AI developer job training track based on industry needs. The characteristic of this curriculum is that it can minimize the mismatch of AI developer job competency between industries and universities. To develop an AI developer job training track, a survey will be conducted for AI developers working in industrial fields. In this survey, among the five NCS-based AI developer jobs, job analysis is conducted by deriving AI developer jobs with high demand for manpower in industrial fields. In job analysis, the core competency unit elements of the job are selected, and knowledge, skills, tools, etc. necessary to perform the core competency unit elements are derived. In addition, a standard AI developer job curriculum is developed by deriving core subjects and road-map that can educate knowledge, skills, tools, etc. In addition, we present an efficient AI developer job training method using the standard AI developer job training course proposed in this paper.

• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.39 no.3
• /
• pp.141-148
• /
• 2021

This paper presents a method to detect zebra-crossing using deep learning which combines SegNet and ResNet. For the blind, a safe crossing system is important to know exactly where the zebra-crossings are. Zebra-crossing detection by deep learning can be a good solution to this problem and robotic vision-based assistive technologies sprung up over the past few years, which focused on specific scene objects using monocular detectors. These traditional methods have achieved significant results with relatively long processing times, and enhanced the zebra-crossing perception to a large extent. However, running all detectors jointly incurs a long latency and becomes computationally prohibitive on wearable embedded systems. In this paper, we propose a model for fast and stable segmentation of zebra-crossing from captured images. The model is improved based on a combination of SegNet and ResNet and consists of three steps. First, the input image is subsampled to extract image features and the convolutional neural network of ResNet is modified to make it the new encoder. Second, through the SegNet original up-sampling network, the abstract features are restored to the original image size. Finally, the method classifies all pixels and calculates the accuracy of each pixel. The experimental results prove the efficiency of the modified semantic segmentation algorithm with a relatively high computing speed.

• Journal of KIISE:Computer Systems and Theory
• /
• v.35 no.5
• /
• pp.213-222
• /
• 2008

This paper is related to color image segmentation and textile texture mapping for the 2D virtual wearing system. The proposed system is characterized as virtually wearing a new textile pattern selected by user to the clothing shape section, based on its intensity difference map, segmented from a 2D clothes model image using color image segmentation technique. Regardless of color or intensity of model clothes, the proposed system is possible to virtually change the textile pattern or color with holding the illumination and shading properties of the selected clothing shape section, and also to quickly and easily simulate, compare, and select multiple textile pattern combinations for individual styles or entire outfits. The proposed system can provide higher practicality and easy-to-use interface, as it makes real-time processing possible in various digital environment, and creates comparatively natural and realistic virtual wearing styles, and also makes semi-automatic processing possible to reduce the manual works to a minimum. According to the proposed system, it can motivate the creative activity of the designers with simulation results on the effect of textile pattern design on the appearance of clothes without manufacturing physical clothes and, as it can help the purchasers for decision-making with them, promote B2B or B2C e-commerce.