• Journal of KIISE:Software and Applications
• /
• v.37 no.8
• /
• pp.629-640
• /
• 2010

The exemplar-based inpainting model is widely used to remove objects from natural images and to restore a damaged region. This paper presents a method which improves the performance of the conventional exemplar-based inpainting model by modifying three major parts in the model: data term, confidence term and patch selection. While the conventional data term is calculated using the local gradient, the proposed method uses 16 compass masks to get the global gradient to make the method robust to noise. To overcome the problem that the confidence term gets negligible in the inside of the eliminated region, a method is proposed which makes the confidence term decrease slowly in the eliminated region. The patch selection procedure is modified so that the closer patch has higher weight. Experiments showed that the proposed method produced more natural images and lower reconstruction error than the conventional exemplar-based inpainting.

• KIISE Transactions on Computing Practices
• /
• v.21 no.12
• /
• pp.745-750
• /
• 2015

LiDAR(Light Detection and Ranging) technology provides realistic 3-dimension image information, and it has been widely utilized in various fields. However, the utilization of this technology in the military domain requires prompt responses to dynamically changing tactical environment and is therefore limited since LiDAR technology requires complex processing in order for extensive amounts of LiDAR data to be utilized. In this paper, we introduce an Unmanned Aircraft Platform Based Real-time LiDAR Data Processing Architecture that can provide real-time detection information by parallel processing and off-loading between the UAV processing and high-performance data processing areas. We also conducted experiments to verify the feasibility of our proposed architecture. Processing with ARM cluster similar to the UAV platform processing area results in similar or better performance when compared to the current method. We determined that our proposed architecture can be utilized in the military domain for tactical and combat purposes such as unmanned monitoring system.

• Journal of KIISE:Software and Applications
• /
• v.30 no.5_6
• /
• pp.444-453
• /
• 2003

Recently, a great interest of bio-technology(BT) is concentrated and the image analysis technique for electrophoresis gels is highly requested to analyze genetic information or to look for some new bio-activation materials. For this purpose, the location and quantity of each band in a lane should be measured. In most of existing techniques, the approach of peak searching in a profile of a lane is used. But this peak is improper as the representative of a band, because its location does not correspond to that of the brightest pixel or the center of gravity. Also, it is improper to measure band quantity in most of these approaches because various enhancement processes are commonly applied to original images to extract peaks easily. In this paper, we adopt an approach to measure accumulated brightness as a band quantity in each band region, which Is extracted by not using any process of changing relative brightness, and the gravity center of the region is calculated as a band location. Actually, we first extract lanes with an entropy-based threshold calculated on a gel-image histogram. And then, three other methods are proposed and applied to extract bands. In the MER method, peaks and valleys are searched on a vertical search line by which each lane is bisected. And the minimum enclosing rectangle of each band is set between successive two valleys. On the other hand, in the RG-1 method, each band is extracted by using region growing with a peak as a seed, separating overlapped neighbor bands. In the RG-2 method, peaks and valleys are searched on two vertical lines by which each lane is trisected, and the left and right peaks nay be paired up if they seem to belong to the same band, and then each band region is grown up with a peak or both peaks if exist. To compare above three methods, we have measured the location and amount of bands. As a result, the average errors in band location of MER, RG-1, and RG-2 were 6%, 3%, and 1%, respectively, when the lane length is normalized to a unit value. And the average errors in band amount were 8%, 5%, and 2%, respectively, when the sum of band amount is normalized to a unit value. In conclusion, RG-2 was shown to be more reliable in the accuracy of measuring the location and amount of bands.

• Journal of KIISE:Computing Practices and Letters
• /
• v.8 no.4
• /
• pp.479-488
• /
• 2002

In recent years the amount of digital video used has risen dramatically to keep pace with the increasing use of the Internet and consequently an automated method is needed for indexing digital video databases. Textual information, both superimposed and embedded scene texts, appearing in a digital video can be a crucial clue for helping the video indexing. In this paper, a new method is presented to extract both superimposed and embedded scene texts in a freeze-frame of news video. The algorithm is summarized in the following three steps. For the first step, a color image is converted into a gray-level image and applies contrast stretching to enhance the contrast of the input image. Then, a modified local adaptive thresholding is applied to the contrast-stretched image. The second step is divided into three processes: eliminating text-like components by applying erosion, dilation, and (OpenClose+CloseOpen)/2 morphological operations, maintaining text components using (OpenClose+CloseOpen)/2 operation with a new Geo-correction method, and subtracting two result images for eliminating false-positive components further. In the third filtering step, the characteristics of each component such as the ratio of the number of pixels in each candidate component to the number of its boundary pixels and the ratio of the minor to the major axis of each bounding box are used. Acceptable results have been obtained using the proposed method on 300 news images with a recognition rate of 93.6%. Also, my method indicates a good performance on all the various kinds of images by adjusting the size of the structuring element.

• Journal of KIISE:Computer Systems and Theory
• /
• v.27 no.8
• /
• pp.691-700
• /
• 2000

The standard volume ray-tracing, optimized with octree, needs to repeatedly traverse hierarchical structures for each ray that often leads to redundant computations. It also employs the expensive 3D interpolation for producing high quality images. In this paper, we present a new ray-casting method that efficiently computes shaded colors and opacities at resampling points by traversing octree only once. This method traverses volume data in object-order, finds resampling points on slices incrementally, and performs resampling based on 2D interpolation. While the early ray-termination, which is one of the most effective optimization techniques, is not easily combined with object-order methods, we solved this problem using a dynamic data structure in image space. Considering that our new method is easy to implement, and need little additional memory, it will be used as very effective volume method that fills the performance gap between ray-casting and shear-warping.

• Journal of KIISE:Software and Applications
• /
• v.27 no.4
• /
• pp.402-411
• /
• 2000

It is an important step in video indexing and retrieval to detect shot boundaries on video data. Some approaches are proposed to detect shot changes by computing color histogram differences or the variances of DCT coefficients. However, these approaches do not consider the content or meaningful features in the image data which are useful in high level video processing. In particular, it is desirable to detect these features from compressed video data because this requires less processing overhead. In this paper, we propose a new method to detect shot boundaries from MPEG data using region flow and color information. First, we reconstruct DC images and compute region flow information and color histogram differences from HSV quantized images. Then, we compute the points at which region flow has discontinuities or color histogram differences are high. Finally, we decide those points as shot boundaries according to our proposed algorithm.

• KIISE Transactions on Computing Practices
• /
• v.20 no.12
• /
• pp.670-675
• /
• 2014

Graphical user interfaces (GUI) in modern software deliver information visually, and a well-designed interface can provide information to the use in an organized and intuitive manner while poorly-designed interfaces can cause visual inconvenience and confusion. In order to effectively deliver information to the user, visual attention should be placed on a prominent location in the image. This paper introduces a method based on a human visual system (HVS) that can improve Metro-style GUIs by reducing a user's workload to visually find information. Our method is designed with spatial mapping and color mapping for buttons in the Metro-style GUI. Also we define a metric for Metro-style GUI effectiveness, including an optimization algorithm. The results show that our method improves the performance of visual search tasks in a Metro-style GUI.

• Journal of KIISE
• /
• v.45 no.1
• /
• pp.22-29
• /
• 2018

Over the decades, neural networks have been successfully used in numerous applications from speech recognition to image classification. However, these neural networks cannot explain their results and one needs to know how and why a specific conclusion was drawn. Most studies focus on extracting binary rules from neural networks, which is often impractical to do, since data sets used for machine learning applications contain continuous values. To fill the gap, this paper presents an algorithm to extract logic rules from a trained neural network for data with continuous attributes. It uses hyperplane-based linear classifiers to extract rules with numeric values from trained weights between input and hidden layers and then combines these classifiers with binary rules learned from hidden and output layers to form non-linear classification rules. Experiments with different datasets show that the proposed approach can accurately extract logical rules for data with nonlinear continuous attributes.

• Journal of KIISE:Software and Applications
• /
• v.34 no.12
• /
• pp.1045-1055
• /
• 2007

Latent topic models are statistical models which automatically captures salient patterns or correlation among features underlying a data collection in a probabilistic way. They are gaining an increased popularity as an effective tool in the application of automatic semantic feature extraction from text corpus, multimedia data analysis including image data, and bioinformatics. Among the important issues for the effectiveness in the application of latent topic models to the massive data set is the efficient learning of the model. The paper proposes an accelerated learning technique for PLSA model, one of the popular latent topic models, by an incremental EM algorithm instead of conventional EM algorithm. The incremental EM algorithm can be characterized by the employment of a series of partial E-steps that are performed on the corresponding subsets of the entire data collection, unlike in the conventional EM algorithm where one batch E-step is done for the whole data set. By the replacement of a single batch E-M step with a series of partial E-steps and M-steps, the inference result for the previous data subset can be directly reflected to the next inference process, which can enhance the learning speed for the entire data set. The algorithm is advantageous also in that it is guaranteed to converge to a local maximum solution and can be easily implemented just with slight modification of the existing algorithm based on the conventional EM. We present the basic application of the incremental EM algorithm to the learning of PLSA and empirically evaluate the acceleration performance with several possible data partitioning methods for the practical application. The experimental results on a real-world news data set show that the proposed approach can accomplish a meaningful enhancement of the convergence rate in the learning of latent topic model. Additionally, we present an interesting result which supports a possible synergistic effect of the combination of incremental EM algorithm with parallel computing.

• Journal of KIISE:Computing Practices and Letters
• /
• v.7 no.5
• /
• pp.416-428
• /
• 2001

In order to build a news archive system. the news video stream should be first segmented into several articles, ad their contents are abstracted effectively. This abstraction helps the users to understand the contents of the article without playing the whole video stream. This paper proposes a new article boundary detection scheme for the news video streams together with a new news article abstraction scheme using the shot types of the news video data. The shots in the news video are classified into anchor person shots, interview shots, speech shots, reporting shots, graphic shots, and others. Since the news article starts with an anchor shot whose duration is relatively longer than other shots with special screen structure, the article boundary in detected by the computing the length of the shot and checking the screen structure in the proposed scheme. For the effective abstraction of the article video, the graphic image located in the right-top of the anchor shot frames is primarily used in the proposed abstraction scheme since it is the abstraction of the article made by the producer of the news according to its contents so that it contains a lot of meaningful information. The key frames of the other shots except interview and report shots are also used to abstract the contents of the articles in the proposed scheme. Upon experimental results, the precision and recall values of the proposed article boundary detection scheme could be 92% and 96%, respectively. This paper also presents a design and implementation of a prototype news archive system on WWW that consists of an indexing tool, an authoring tool, a database for meta-data of the news, and a browsing tool.