• Proceedings of the KSRS Conference
• /
• 1999.11a
• /
• pp.405-410
• /
• 1999

Topographic maps can be made and updated with satellite images, but it requires many human interactions that are inefficient and costly. Therefore, the automatizing of the road extraction procedures could increase efficiency in terms of time and cost. Although methods of extracting roads, railroads and rivers from satellite images have been developed in many studies, studies on the road extraction from satellite images of urbanized area are still not relevant, because many artificial components In the city makes the delineation of the roads difficult. So, to extract roads from high resolution satellite images of urbanized area, this study has proposed the combined use of wavelet transform and multi-resolution analysis. In consequence, this study verifies that it is possible to automatize the road extraction from satellite images of urbanized area. And to realize the automatization more completely, various algorithms need to be developed.

• Journal of The Korean Society of Agricultural Engineers
• /
• v.54 no.2
• /
• pp.47-53
• /
• 2012

Recently many developed countries have used satellite images for classifying cropland areas to reduce time and efforts put into field survey. Korea also has used satellite images for the same purpose since KOMPSAT-2 was successfully launched and operated in 2006, but still far way to go in order to achieve the required accuracy from the products. This study evaluated the accuracy of the calculated croplands by using the objected classification method with various satellite images including ASTER, Spot-5, Rapid eye, Quickbird-2, Geo eye-1. Also, their usability and effectiveness for the cropland survey were verified by comparing with field survey data. As results. Geo eye-1 and Rapid eye showed higher accuracy to calculate the paddy field areas while Geo eye-1 and Quickbird-2 showed higher accuracy to calculate the upland field areas.

• Proceedings of the Korea Multimedia Society Conference
• /
• 2012.05a
• /
• pp.205-208
• /
• 2012

Nowadays GPU (Graphic Process Unit) is not only used to show and render some images, but also for another computation. In this paper, we tried to use GPU to do some morphology operations to remove skull from axial MRI images. This skull removing process is an important step in brain segmentation because we would like to work with the brain only, without any skull on it. The result shows that simple morphology operations to remove skull has been successfully applied on MRI images, but there are still many parts that can be develop to get better images.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 1996.06b
• /
• pp.50-55
• /
• 1996

Multimedia boom has happened worldwide these days. In multimedia, we use several kinds of media such as character, figure, voice, music, still images, moving picture etc.. Then I think image including moving picture is the most effective and important media for human being. Creating digital images using a computer has the following two main approaches, depending on how the computer is used. 1. CG Technology. Created images, produced through computer graphics. 2. Digital Image Processing. Images processed through digital image processing technologies. Approach (1) is very popular as Computer Graphics. Two-dimensional and three-dimensional computer graphics techniques are used over wide applications today. On the other hand, Approach (2), which uses digital image processing technology, has been attracting attention lately, in the filed of movies and television. In this report, I will introduce these approaches of CG and digital image processing, and show some application fields such as current movies.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 1996.06b
• /
• pp.29-34
• /
• 1996

This paper proposes a new concept in video called Graphical Video. Graphical Video is a content-based and scalable video representation. A video consists of several elements such as moving images, still images, graphics, characters and charts. All of these elements can be represented graphically except moving images. It is desirable to transform these moving images graphical elements so that they can be treated in the same way as other graphical elements. To achieve this, we propose a new graphical representation of moving images using spatio-temporal clusters, which consist of texture and contours. The texture is described by three-dimensional fractal coefficients, while the contours are described by polygons. We propose a method that gives domain pool location and size as a means to describe cluster texture within or near a region of clusters. Results of an experiment on texture quality confirm that the method provides sufficiently high SNR as compared to that in the original three-dimensional fractal approximation.

• The Transactions of the Korean Institute of Electrical Engineers D
• /
• v.52 no.3
• /
• pp.148-155
• /
• 2003

In this paper, a nonlinear anisotropic filtering method without the loss of important information happened due to the repeated filtering in magnetic resonance images is proposed. First of all original images are divided into four regions, e.g., SPR(Strong Plain Region), EPR(Easy Plain Region), SER(Strong Edge Region), and EER(Easy Edge Region). An optimal template among multiple templates is selected, then the nonlinear anisotropic filtering based on the template is applied in pixel by pixel basis. In the proposed algorithm, filtering strength of EER containing important information is adjusted very weak and filtering strength for remaining regions is also adjusted according to the degree of the importance. In spite of repeated filtering, resulting images by the proposed method could still preserve anatomy information of original images without any degradation. Compared to the existing nonlinear anisotropic filtering, the proposed filtering method with multiple templates provides higher reliability for filtered images.

• Transactions on Electrical and Electronic Materials
• /
• v.14 no.4
• /
• pp.187-192
• /
• 2013

Because of the importance of small target detection in infrared images, many studies have been carried out in this area. Using a Kalman filter and mean shift algorithm, this study proposes an algorithm to track multiple small moving targets even in cases of target disappearance and appearance in serial infrared images in an environment with many noises. Difference images, which highlight the background images estimated with a background estimation filter from the original images, have a relatively very bright value, which becomes a candidate target area. Multiple target tracking consists of a Kalman filter section (target position prediction) and candidate target classification section (target selection). The system removes error detection from the detection results of candidate targets in still images and associates targets in serial images. The final target detection locations were revised with the mean shift algorithm to have comparatively low tracking location errors and allow for continuous tracking with standard model updating. In the experiment with actual marine infrared serial images, the proposed system was compared with the Kalman filter method and mean shift algorithm. As a result, the proposed system recorded the lowest tracking location errors and ensured stable tracking with no tracking location diffusion.

• Proceedings of the IEEK Conference
• /
• 2002.06d
• /
• pp.355-358
• /
• 2002

In this paper, we propose RTP(Real-Time Transport Protocol) based image transport system to transport still images in real-time after JPEG2000 compression, which is still image compression standard for next generation. To add RTP packet on UDP packet, the image transport system inserts packetizer and depacketizer process into transmitter and receiver of RTP data, respectively. We apply the proposed system to several image and compare the transport time to TCP-based method.

• Design & Manufacturing
• /
• v.15 no.3
• /
• pp.1-6
• /
• 2021

An artificial neural network model based on a deep learning algorithm is known to be more accurate than humans in image classification, but there is still a limit in the sense that there needs to be a lot of training data that can be called big data. Therefore, various techniques are being studied to build an artificial neural network model with high precision, even with small data. The transfer learning technique is assessed as an excellent alternative. As a result, the purpose of this study is to develop an artificial neural network system that can classify burr images of light guide plate products with 99% accuracy using transfer learning technique. Specifically, for the light guide plate product, 150 images of the normal product and the burr were taken at various angles, heights, positions, etc., respectively. Then, after the preprocessing of images such as thresholding and image augmentation, for a total of 3,300 images were generated. 2,970 images were separated for training, while the remaining 330 images were separated for model accuracy testing. For the transfer learning, a base model was developed using the NASNet-Large model that pre-trained 14 million ImageNet data. According to the final model accuracy test, the 99% accuracy in the image classification for training and test images was confirmed. Consequently, based on the results of this study, it is expected to help develop an integrated AI production management system by training not only the burr but also various defective images.

• Journal of Korea Multimedia Society
• /
• v.24 no.12
• /
• pp.1624-1631
• /
• 2021

With the continued development of image processing technology, we live in a time when it is difficult to visually discriminate processed (or tampered) images from real images. However, as the risk of fake images being misused for crime increases, the importance of image forensic science for identifying fake images is emerging. Currently, various deep learning-based identifiers have been studied, but there are still many problems to be used in real situations. Due to the inherent characteristics of deep learning that strongly relies on given training data, it is very vulnerable to evaluating data that has never been viewed. Therefore, we try to find a way to improve generalization ability of deep learning-based fake image identifiers. First, images with various contents were added to the training dataset to resolve the over-fitting problem that the identifier can only classify real and fake images with specific contents but fails for those with other contents. Next, color spaces other than RGB were exploited. That is, fake image identification was attempted on color spaces not considered when creating fake images, such as HSV and YCbCr. Finally, dropout, which is commonly used for generalization of neural networks, was used. Through experimental results, it has been confirmed that the color space conversion to HSV is the best solution and its combination with the approach of increasing the training dataset significantly can greatly improve the accuracy and generalization ability of deep learning-based identifiers in identifying fake images that have never been seen before.