• ETRI Journal
• /
• v.29 no.5
• /
• pp.700-702
• /
• 2007

We present a relevance feedback approach based on multi-class support vector machine (SVM) learning and cluster-merging which can significantly improve the retrieval performance in region-based image retrieval. Semantically relevant images may exhibit various visual characteristics and may be scattered in several classes in the feature space due to the semantic gap between low-level features and high-level semantics in the user's mind. To find the semantic classes through relevance feedback, the proposed method reduces the burden of completely re-clustering the classes at iterations and classifies multiple classes. Experimental results show that the proposed method is more effective and efficient than the two-class SVM and multi-class relevance feedback methods.

• Journal of Broadcast Engineering
• /
• v.27 no.5
• /
• pp.808-811
• /
• 2022

Display under strong ambient lighting is perceived darker than it really is. Existing techniques for solving the problem in terms of software show limitations in that image enhancement techniques are applied regardless of ambient lighting or chrominance is not improved compared to luminance. Therefore, this paper proposes a visibility enhancement algorithm using deep learning to adaptively respond to ambient lighting values and an equation to restore optimal chrominance for luminance. The algorithm receives an ambient lighting value with the input image, and then applies a deep learning model and chrominance restoration equation to generate an image to minimize the difference between the degradation modeling of enhanced image and the input image. Qualitative evaluation proves that the algorithm shows excellent performance in improving visibility under strong ambient lighting through comparison of images applied with degradation modeling.

• The Journal of Korea Robotics Society
• /
• v.14 no.1
• /
• pp.14-21
• /
• 2019

In the ground environment, mobile robot research uses sensors such as GPS and optical cameras to localize surrounding landmarks and to estimate the position of the robot. However, an underwater environment restricts the use of sensors such as optical cameras and GPS. Also, unlike the ground environment, it is difficult to make a continuous observation of landmarks for location estimation. So, in underwater research, artificial markers are installed to generate a strong and lasting landmark. When artificial markers are acquired with an underwater sonar sensor, different types of noise are caused in the underwater sonar image. This noise is one of the factors that reduces object detection performance. This paper aims to improve object detection performance through distortion and rotation augmentation of training data. Object detection is detected using a Faster R-CNN.

• The Journal of Korean Institute of Next Generation Computing
• /
• v.15 no.5
• /
• pp.64-74
• /
• 2019

Although a non-rigid registration has high demands in clinical practice, it has a high computational complexity and it is very difficult for ensuring the accuracy and robustness of registration. This study proposes a method of applying a non-rigid registration to 3D magnetic resonance images of brain in an unsupervised learning environment by using a deep-learning network. A feature vector between two images is produced through the network by receiving both images from two different patients as inputs and it transforms the target image to match the source image by creating a displacement vector field. The network is designed based on a U-Net shape so that feature vectors that consider all global and local differences between two images can be constructed when performing the registration. As a regularization term is added to a loss function, a transformation result similar to that of a real brain movement can be obtained after the application of trilinear interpolation. This method enables a non-rigid registration with a single-pass deformation by only receiving two arbitrary images as inputs through an unsupervised learning. Therefore, it can perform faster than other non-learning-based registration methods that require iterative optimization processes. Our experiment was performed with 3D magnetic resonance images of 50 human brains, and the measurement result of the dice similarity coefficient confirmed an approximately 16% similarity improvement by using our method after the registration. It also showed a similar performance compared with the non-learning-based method, with about 10,000 times speed increase. The proposed method can be used for non-rigid registration of various kinds of medical image data.

• Journal of Convergence for Information Technology
• /
• v.8 no.6
• /
• pp.375-380
• /
• 2018

This paper proposes a classification method using the Convolutional Neural Network(CNN) which can obtain the type of trucks from the input image without the feature extraction step. To automatically classify vehicle images according to the type of truck cargo box, the top view images of the vehicle are used as input image and we design the structure of the CNN suitable for the input images. Learning images and correct output results is generated and the weights of neural network are obtained through the learning process. The actual image is input to the CNN and the output of the CNN is calculated. The classification performance is evaluated through comparison CNN output with actual vehicle types. Experimental results show that vehicle images could be classified with more than 90 percent accuracy according to the type of cargo box and this method can be used for pre-classification for inspecting loading defect.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.15 no.4
• /
• pp.709-716
• /
• 2020

In social media, when posting a post, tag information of an image is generally used because the search is mainly performed using a tag. Users want to expose the post to many people by attaching the tag to the post. Also, the user has trouble posting the tag to be tagged along with the post, and posts that have not been tagged are also posted. In this paper, we propose a method to find an image similar to the input image, extract the label attached to the image, find the posts on instagram, where the label exists as a tag, and recommend other tags in the post. In the proposed method, the label is extracted from the image through the model of the convolutional neural network (CNN) deep learning technique, and the instagram is crawled with the extracted label to sort and recommended tags other than the label. We can see that it is easy to post an image using the recommended tag, increase the exposure of the search, and derive high accuracy due to fewer search errors.

• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.39 no.6
• /
• pp.437-456
• /
• 2021

In the field of remote sensing in Korea, starting in 2017, deep learning has begun to show efficient research results compared to existing research methods. Currently, research is being conducted to apply deep learning in almost all fields of remote sensing, from image preprocessing to applications. To analyze the research trend of deep learning applied to the remote sensing field, Korean domestic journal papers, published until October 2021, related to deep learning applied to the remote sensing field were collected. Based on the collected 60 papers, research trend analysis was performed while focusing on deep learning network purpose, remote sensing application field, and remote sensing image acquisition platform. In addition, open source data that can be effectively used to build training data for performing deep learning were summarized in the paper. Through this study, we presented the problems that need to be solved in order for deep learning to be established in the remote sensing field. Moreover, we intended to provide help in finding research directions for researchers to apply deep learning technology into the remote sensing field in the future.

• Journal of Information Processing Systems
• /
• v.19 no.1
• /
• pp.67-79
• /
• 2023

In the image field, data augmentation refers to increasing the amount of data through an editing method such as rotating or cropping a photo. In this study, a generative adversarial network (GAN) image was created using CycleGAN, and various colors of dogs were reflected through data augmentation. In particular, dog data from the Stanford Dogs Dataset and Oxford-IIIT Pet Dataset were used, and 10 breeds of dog, corresponding to 300 images each, were selected. Subsequently, a GAN image was generated using CycleGAN, and four learning groups were established: 2,000 original photos (group I); 2,000 original photos + 1,000 GAN images (group II); 3,000 original photos (group III); and 3,000 original photos + 1,000 GAN images (group IV). The amount of data in each learning group was augmented using existing data augmentation methods such as rotating, cropping, erasing, and distorting. The augmented photo data were used to train the MobileNet_v3_Large, ResNet-152, InceptionResNet_v2, and NASNet_Large frameworks to evaluate the classification accuracy and loss. The top-3 accuracy for each deep neural network model was as follows: MobileNet_v3_Large of 86.4% (group I), 85.4% (group II), 90.4% (group III), and 89.2% (group IV); ResNet-152 of 82.4% (group I), 83.7% (group II), 84.7% (group III), and 84.9% (group IV); InceptionResNet_v2 of 90.7% (group I), 88.4% (group II), 93.3% (group III), and 93.1% (group IV); and NASNet_Large of 85% (group I), 88.1% (group II), 91.8% (group III), and 92% (group IV). The InceptionResNet_v2 model exhibited the highest image classification accuracy, and the NASNet_Large model exhibited the highest increase in the accuracy owing to data augmentation.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.16 no.6
• /
• pp.1231-1238
• /
• 2021

Recently, large online shopping malls are providing image search services as well as text or category searches. However, in the case of an image search service, there is a problem in that the search service cannot be used in the absence of an image. This paper describes the development of a system that allows users to find the clothes they want through hand-drawn images of the style of clothes when they search for clothes in an online clothing shopping mall. The hand-drawing data drawn by the user increases the accuracy of matching through neural network learning, and enables matching of clothes using various object detection algorithms. This is expected to increase customer satisfaction with online shopping by allowing users to quickly search for clothing they are looking for.

• Journal of IKEEE
• /
• v.26 no.2
• /
• pp.176-185
• /
• 2022

Data augmentation is general approach to solve overfitting of machine learning models by applying various data transformations and distortions to dataset. However, when data augmentation is applied in GAN-based model, which is deep learning image generation model, data transformation and distortion are reflected in the generated image, then the generated image quality decrease. To prevent this problem called augmentation leak, we propose a scheme that can prevent augmentation leak regardless of the type and number of augmentations. Specifically, we analyze the conditions of augmentation leak occurrence by type and implement auxiliary augmentation task classifier that can prevent augmentation leak. Through experiments, we show that the proposed technique prevents augmentation leak in the GAN model, and as a result improves the quality of the generated image. We also demonstrate the superiority of the proposed scheme through ablation study and comparison with other representative augmentation leak prevention technique.