• Science of Emotion and Sensibility
• /
• v.7 no.4
• /
• pp.57-65
• /
• 2004

In this paper, we present a hybrid approach which incorporates color, texture and shape in content-based image retrieval. Colors in each image are clustered into a small number of representative colors. The feature descriptor consists of the representative colors and their percentages in the image. A similarity measure similar to the cumulative color histogram distance measure is defined for this descriptor. The co-occurrence matrix as a statistical method is used for texture analysis. An optimal set of five statistical functions are extracted from the co-occurrence matrix of each image, in order to render the feature vector for eachimage maximally informative. The edge information captured within edge histograms is extracted after a pre-processing phase that performs color transformation, quantization, and filtering. The features where thus extracted and stored within feature vectors and were later compared with an intersection-based method. The content-based retrieval system is tested to be effective in terms of retrieval and scalability through experimental results and precision-recall analysis.

• Korean Journal of Remote Sensing
• /
• v.34 no.5
• /
• pp.739-747
• /
• 2018

Integration analysis of multi-sensor satellite images is becoming increasingly important. The first step in integration analysis is image registration between multi-sensor. SIFT (Scale Invariant Feature Transform) is a representative image registration method. However, optical image and SAR (Synthetic Aperture Radar) images are different from sensor attitude and radiation characteristics during acquisition, making it difficult to apply the conventional method, such as SIFT, because the radiometric characteristics between images are nonlinear. To overcome this limitation, we proposed a modified method that combines the SAR-SIFT method and shape descriptor vector DLSS(Dense Local Self-Similarity). We conducted an experiment using two pairs of Cosmo-SkyMed and KOMPSAT-2 images collected over Daejeon, Korea, an area with a high density of buildings. The proposed method extracted the correct matching points when compared to conventional methods, such as SIFT and SAR-SIFT. The method also gave quantitatively reasonable results for RMSE of 1.66m and 2.45m over the two pairs of images.

• Journal of the Korea Convergence Society
• /
• v.11 no.3
• /
• pp.19-26
• /
• 2020

In this paper, we propose a method to analyze the loss region of the dictionary-based super resolution result learned for image quality improvement and to map the learning data according to the analyzed loss region. In the conventional learned dictionary-based method, a result different from the feature configuration of the input image may be generated according to the learning image, and an unintended artifact may occur. The proposed method estimate loss information of low resolution images by analyzing the reconstructed contents to reduce inconsistent feature composition and unintended artifacts in the example-based super resolution process. By mapping the training data according to the final interpolation feature map, which improves the noise and pixel imbalance of the estimated loss information using a Gaussian-based kernel, it generates super resolution with improved noise, artifacts, and staircase compared to the existing super resolution. For the evaluation, the results of the existing super resolution generation algorithms and the proposed method are compared with the high-definition image, which is 4% better in the PSNR (Peak Signal to Noise Ratio) and 3% in the SSIM (Structural SIMilarity Index).

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.29 no.4
• /
• pp.775-784
• /
• 2019

Malware infringement attacks are continuously increasing in various environments such as mobile, IOT, windows and mac due to the emergence of new and variant malware, and signature-based countermeasures have limitations in detection of malware. In addition, analytical performance is deteriorating due to obfuscation, packing, and anti-VM technique. In this paper, we propose a system that can detect malware based on machine learning by using similarity hashing-based pattern detection technique and static analysis after file classification according to packing. This enables more efficient detection because it utilizes both pattern-based detection, which is well-known malware detection, and machine learning-based detection technology, which is advantageous for detecting new and variant malware. The results of this study were obtained by detecting accuracy of 95.79% or more for benign sample files and malware sample files provided by the AI-based malware detection track of the Information Security R&D Data Challenge 2018 competition. In the future, it is expected that it will be possible to build a system that improves detection performance by applying a feature vector and a detection method to the characteristics of a packed file.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.17 no.3
• /
• pp.881-895
• /
• 2023

Modern image inpainting techniques based on deep learning have achieved remarkable performance, and more and more people are working on repairing more complex and larger missing areas, although this is still challenging, especially for facial image inpainting. For a face image with a huge missing area, there are very few valid pixels available; however, people have an ability to imagine the complete picture in their mind according to their subjective will. It is important to simulate this capability while maintaining the identity features of the face as much as possible. To achieve this goal, we propose a three-stage network model, which we refer to as the identity and structure feature refinement network (ISFRNet). ISFRNet is based on 1) a pre-trained pSp-styleGAN model that generates an extremely realistic face image with rich structural features; 2) a shallow structured network with a small receptive field; and 3) a modified U-net with two encoders and a decoder, which has a large receptive field. We choose structural similarity index (SSIM), peak signal-to-noise ratio (PSNR), L1 Loss and learned perceptual image patch similarity (LPIPS) to evaluate our model. When the missing region is 20%-40%, the above four metric scores of our model are 28.12, 0.942, 0.015 and 0.090, respectively. When the lost area is between 40% and 60%, the metric scores are 23.31, 0.840, 0.053 and 0.177, respectively. Our inpainting network not only guarantees excellent face identity feature recovery but also exhibits state-of-the-art performance compared to other multi-stage refinement models.

• Journal of Korea Multimedia Society
• /
• v.7 no.2
• /
• pp.173-186
• /
• 2004

This paper proposed the content-based retrieval system as a method for performing image retrieval through the effective feature extraction of the object of significant meaning based on the characteristics of man's visual system. To allow the object region of interest to be primarily detected, the region, being comparatively large size, greatly different from the background color and located in the middle of the image, was judged as the major object with a meaning. To get the original features of the image, the cumulative sum of tile declination difference vector the segment of the object contour had and the signature of the bipartite object were extracted and used in the form of being applied to the rotation of the object and the change of the size after partition of the total length of the object contour of the image into the normalized segment. Starting with this form feature, it was possible to make a retrieval robust to any change in translation, rotation and scaling by combining information on the texture sample, color and eccentricity and measuring the degree of similarity. It responded less sensitively to the phenomenon of distortion of the object feature due to the partial change or damage of the region. Also, the method of imposing a different weight of similarity on the image feature based on the relationship of complexity between measured objects using the fractal dimension by the Boxing-Counting Dimension minimized the wrong retrieval and showed more efficient retrieval rate.

• Journal of the Korea Convergence Society
• /
• v.12 no.1
• /
• pp.99-104
• /
• 2021

In this paper, we propose health risk management using feature extraction and cluster analysis considering time flow. The proposed method proceeds in three steps. The first is the pre-processing and feature extraction step. It collects user's lifelog using a wearable device, removes incomplete data, errors, noise, and contradictory data, and processes missing values. Then, for feature extraction, important variables are selected through principal component analysis, and data similar to the relationship between the data are classified through correlation coefficient and covariance. In order to analyze the features extracted from the lifelog, dynamic clustering is performed through the K-means algorithm in consideration of the passage of time. The new data is clustered through the similarity distance measurement method based on the increment of the sum of squared errors. Next is to extract information about the cluster by considering the passage of time. Therefore, using the health decision-making system through feature clusters, risks able to managed through factors such as physical characteristics, lifestyle habits, disease status, health care event occurrence risk, and predictability. The performance evaluation compares the proposed method using Precision, Recall, and F-measure with the fuzzy and kernel-based clustering. As a result of the evaluation, the proposed method is excellently evaluated. Therefore, through the proposed method, it is possible to accurately predict and appropriately manage the user's potential health risk by using the similarity with the patient.

• The KIPS Transactions:PartB
• /
• v.16B no.3
• /
• pp.233-238
• /
• 2009

This paper presents an approach to the classifications of small size document using the instance-based feature filtering Relief-F algorithm. In the document classifications, we have not always good classification performances of small size document included a few features. Because total number of feature in the document set is large, but feature count of each document is very small relatively, so the similarities between documents are very low when we use general assessment of similarity and classifiers. Specially, in the cases of the classification of web document in the directory service and the classification of the sectors that cannot connect with the original file after recovery hard-disk, we have not good classification performances. Thus, we propose the Extended Relief-F(ERelief-F) algorithm using instance-based feature filtering algorithm Relief-F to solve problems of Relief-F as preprocess of classification. For the performance comparison, we tested information gain, odds ratio and Relief-F for feature filtering and getting those feature values, and used kNN and SVM classifiers. In the experimental results, the Extended Relief-F(ERelief-F) algorithm, compared with the others, performed best for all of the datasets and reduced many irrelevant features from document sets.

• The Journal of Korea Institute of Information, Electronics, and Communication Technology
• /
• v.7 no.2
• /
• pp.106-110
• /
• 2014

This paper proposed the face recognition method which can be effectively applied to the robot's vision system. The proposed algorithm is recognition using hue extraction and feature point. hue extraction was using difference of skin color, pupil color, lips color. Features information were extraction from eye, nose and mouth using feature parameters of the difference between the feature point, distance ratio, angle, area. Feature parameters fuzzified data with the data generated by membership function, then evaluate the degree of similarity was the face recognition. The result of experiment are conducted with frontal color images of face as input images the received recognition rate of 96%.

• Journal of Korea Multimedia Society
• /
• v.23 no.2
• /
• pp.127-134
• /
• 2020

In this paper, we propose a map alignment method for maps generated by point-line monocular SLAM. In the proposed method, the information of feature lines as well as feature points extracted from multiple maps are fused into a single map. To this end, the proposed method first searches for similar areas between maps via Bag-of-Words-based image matching. Thereafter, it calculates the similarity transformation between the maps in the corresponding areas to align the maps. Finally, we merge the overlapped information of multiple maps into a single map by removing duplicate information from similar areas. Experimental results show that maps created by different users are combined into a single map, and the accuracy of the fused map is similar with the one generated by a single user. We expect that the proposed method can be utilized for fast imagery map generation.