• Convergence Security Journal
• /
• v.18 no.3
• /
• pp.53-59
• /
• 2018

In this paper, we propose a hierarchical image classification scheme for efficient object image classification. In the non-hierarchical image classification, which classifies the existing whole images at one time, it showed that objects with relatively similar shapes are not recognized efficiently. Therefore, in this paper, we introduce the image classification method in the hierarchical structure which attempts to classify object images hierarchically. Also, we introduce to the efficient class structure and algorithms considering the scalability that can occur when a deep learning image classification is applied to an actual system. Such a scheme makes it possible to classify images with a higher degree of confidence in object images having relatively similar shapes.

• Journal of Korea Multimedia Society
• /
• v.21 no.8
• /
• pp.858-863
• /
• 2018

In this paper, we propose and implement a multi-level convolutional neural network (CNN) algorithm to identify the sexually explicit and lewdness of various image files, and verify its effectiveness by using unauthorized image files generated in the actual military. The proposed algorithm increases the accuracy by applying the convolutional artificial neural network step by step to minimize classification error between similar categories. Experimental data have categorized 20,005 images in the real field into 6 authorization categories and 11 non-authorization categories. Experimental results show that the overall detection rate is 99.51% for the image files. In particular, the excellence of the proposed algorithm is verified through reducing the identification error rate between similar categories by 64.87% compared with the general CNN algorithm.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.10 no.2
• /
• pp.825-836
• /
• 2016

In this paper, a speed-up technique for the non-local means (NLM) image denoising method based on local binary descriptor (LBD) is proposed. In the NLM, most of the computation time is spent on searching for non-local similar patches in the search window. The local binary descriptor which represents the structure of patch as binary strings is employed to speed up the search process in the NLM. The descriptor allows for a fast and accurate preselection of non-local similar patches by bitwise operations. Using this approach, a tradeoff between time-saving and noise removal can be obtained. Simulations exhibit that despite being principally constructed for speed, the proposed algorithm outperforms in terms of denoising quality as well. Furthermore, a parallel implementation on GPU brings NLM-LBD to real-time image denoising.

• Journal of Information Processing Systems
• /
• v.6 no.4
• /
• pp.435-452
• /
• 2010

This paper proposes a novel reversible data hiding scheme based on a Vector Quantization (VQ) codebook. The proposed scheme uses the principle component analysis (PCA) algorithm to sort the codebook and to find two similar codewords of an image block. According to the secret to be embedded and the difference between those two similar codewords, the original image block is transformed into a difference number table. Finally, this table is compressed by entropy coding and sent to the receiver. The experimental results demonstrate that the proposed scheme can achieve greater hiding capacity, about five bits per index, with an acceptable bit rate. At the receiver end, after the compressed code has been decoded, the image can be recovered to a VQ compressed image.

• 제어로봇시스템학회:학술대회논문집
• /
• 2001.10a
• /
• pp.35.5-35
• /
• 2001

The purpose of the study is to develop a synthetic method for generating arbitrary number of not the same but similar texture images. The method includes processes to extract basic shape elements from texture images originating in actual objects, to select them to reappear the image features and to arrange them in a image plane. The authors have already proposed the shape-pass type filter bank assuming that the sensual impression mainly depends on minute shapes existing in the texture images. By use of nine basic shape elements, namely black/white-roof, black/white-line, black/white-snake, black/white-pepper, and cliff, natural texture images originating in actual objects have been characterized by feature vectors in a nine dimensional space. To generate arbitrary number of similar texture images, minute shape pieces ...

• Journal of the Korea Institute of Military Science and Technology
• /
• v.25 no.3
• /
• pp.300-310
• /
• 2022

A vast amount of labeled data is required for deep neural network training. A typical strategy to improve the performance of a neural network given a training data set is to use data augmentation technique. The goal of this work is to offer a novel image augmentation method for improving object detection accuracy. An object in an image is removed, and a similar object from the training data set is placed in its area. An in-painting algorithm fills the space that is eliminated but not filled by a similar object. Our technique shows at most 2.32 percent improvements on mAP in our testing on a military vehicle dataset using the YOLOv4 object detector.

• 제어로봇시스템학회:학술대회논문집
• /
• 2004.08a
• /
• pp.810-813
• /
• 2004

Indexing and Retrieving images from large and varied collections using image content as a key is a challenging and important problem in computer vision application. In this paper, a color Content-based Image Retrieval (CBIR) system using hierarchical Region of Interest (ROI) query and indexing is presented. During indexing process, First, The ROIs on every image in the image database are extracted using a region-based image segmentation technique, The JSEG approach is selected to handle this problem in order to create color-texture regions. Then, Color features in form of histogram and correlogram are then extracted from each segmented regions. Finally, The features are stored in the database as the key to retrieve the relevant images. As in the retrieval system, users are allowed to select ROI directly over the sample or user's submission image and the query process then focuses on the content of the selected ROI in order to find those images containing similar regions from the database. The hierarchical region-of-interest query is performed to retrieve the similar images. Two-level search is exploited in this paper. In the first level, the most important regions, usually the large regions at the center of user's query, are used to retrieve images having similar regions using static search. This ensures that we can retrieve all the images having the most important regions. In the second level, all the remaining regions in user's query are used to search from all the retrieved images obtained from the first level. The experimental results using the indexing technique show good retrieval performance over a variety of image collections, also great reduction in the amount of searching time.

• KIPS Transactions on Software and Data Engineering
• /
• v.10 no.3
• /
• pp.109-114
• /
• 2021

GAN(Generative Adversarial Network) is an image generation model, which is composed of a generator network and a discriminator network, and generates an image similar to a real image. Since the image generated by the GAN should be similar to the actual image, a loss function is used to minimize the loss error of the generated image. However, there is a problem that the loss function of GAN degrades the quality of the image by making the learning to generate the image unstable. To solve this problem, this paper analyzes GAN-related studies and proposes an edge GAN(eGAN) using edge detection. As a result of the experiment, the eGAN model has improved performance over the existing GAN model.

• The Journal of the Korea Contents Association
• /
• v.21 no.8
• /
• pp.1-9
• /
• 2021

Deep learning in computer vision has made accelerated improvement over a short period but large-scale learning data and computing power are still essential that required time-consuming trial and error tasks are involved to derive an optimal network model. In this study, we propose a similar image classification performance improvement method based on CR (Confusion Rate) that considers only the characteristics of the data itself regardless of network optimization or data reinforcement. The proposed method is a technique that improves the performance of the deep learning model by calculating the CRs for images in a dataset with similar characteristics and reflecting it in the weight of the Loss Function. Also, the CR-based recognition method is advantageous for image identification with high similarity because it enables image recognition in consideration of similarity between classes. As a result of applying the proposed method to the Resnet18 model, it showed a performance improvement of 0.22% in HanDB and 3.38% in Animal-10N. The proposed method is expected to be the basis for artificial intelligence research using noisy labeled data accompanying large-scale learning data.

• Journal of radiological science and technology
• /
• v.44 no.5
• /
• pp.443-449
• /
• 2021

In this study, the purpose is to present the foot inclination angle for realizing an image similar to that of the existing examination method and to present the clinical usefulness of the new examination method through comparison between the existing examination method and the newly designed standing foot oblique projection. A foot phantom was used, and the magnification of the image according to the angle was quantitatively evaluated by attaching a nut to the position of the cuboid of the phantom. The internal oblique image acquired using a 30° wedge was set as the standard image. And that image was compared with the images acquired by changing the angle of the foot from 20° to 65° at intervals of 5°. Image evaluation was performed by 3 radiological technologists, and qualitative evaluation using a Likert 5-point scale for evaluation items of true oblique view and quantitative evaluation of the value obtained by measuring the diameter of a nut in each image were performed as image evaluation. For data analysis, reliability analysis between the measure and comparative analysis of the average value for each angle were performed. The qualitative evaluation score for each image was 4.5 to 5 points for most questions in the case of the standard image. And 4 points or less for most questions in the images with a foot angle of 45° or less, and an evaluation score close to the standard image was obtained in the image of 50° or more. And in the quantitative evaluation, the diameter of the nut was measured to be 9.28~9.56 mm. The qualitative evaluation showed a reliability of 0.95~1.0 and the quantitative evaluation was 0.62. As a result of comparing and analyzing the average of the quantitative and qualitative average values for each angle image, the group with the average value most similar to the standard image was images obtained at 55° and 60°, and in the post-analysis, the images of both groups were the same group as the standard image(p<0.01). As a result of this study, it was found that the angle of inclination of the foot for realizing the image most similar to the existing image in the standing foot oblique projection is 55°~60°. In addition, if this test method is applied to the clinic, it is believed that it will help prevent safety accidents such as falls during the test and improve test efficiency by minimizing the movement of patients for the test.