• Journal of IKEEE
• /
• v.28 no.2
• /
• pp.209-220
• /
• 2024

A sketch is an intuitive means to express information, but compared to actual images, it has the problem of being highly abstract, diverse, and sparse. Recent advances in deep learning models have made it possible to discover features that are common to images and sketches. In this paper, we summarize recent trends in sketch-based image retrieval (SBIR) but it is not limited to SBIR. Besides SBIR, we also introduce sketch-based image recognition and generation studies. Zero-shot learning enables models to recognize categories not encountered during training. Zero-shot SBIR methods are also discussed. Commonly used free-hand sketch datasets are summarized and retrieval performance based on these datasets is reported.

• Journal of information and communication convergence engineering
• /
• v.16 no.3
• /
• pp.160-165
• /
• 2018

This paper designed and developed the image processing system of integrating feature extraction and matching by using convolutional neural network (CNN), rather than relying on the simple method of processing feature extraction and matching separately in the image processing of conventional image recognition system. To implement it, the proposed system enables CNN to operate and analyze the performance of conventional image processing system. This system extracts the features of an image using CNN and then learns them by the neural network. The proposed system showed 84% accuracy of recognition. The proposed system is a model of recognizing learned images by deep learning. Therefore, it can run in batch and work easily under any platform (including embedded platform) that can read all kinds of files anytime. Also, it does not require the implementing of feature extraction algorithm and matching algorithm therefore it can save time and it is efficient. As a result, it can be widely used as an image recognition program.

• The Journal of Korea Robotics Society
• /
• v.13 no.1
• /
• pp.1-7
• /
• 2018

The image-to-image translation is one of the deep learning applications using image data. In this paper, we aim at improving the performance of object transfiguration which transforms a specific object in an image into another specific object. For object transfiguration, it is required to transform only the target object and maintain background images. In the existing results, however, it is observed that other parts in the image are also transformed. In this paper, we have focused on the structure of artificial neural networks that are frequently used in the existing methods and have improved the performance by adding constraints to the exiting structure. We also propose the advanced structure that combines the existing structures to maintain their advantages and complement their drawbacks. The effectiveness of the proposed methods are shown in experimental results.

• Journal of Korea Multimedia Society
• /
• v.22 no.6
• /
• pp.646-654
• /
• 2019

Image denoising is considered as a key factor for capturing high-quality photos in digital cameras. Thus far, several image denoising methods have been proposed in the past decade. In addition, previous studies either relied on deep learning-based approaches or used the hand-crafted filters. Unfortunately, the previous method mostly emphasized on image denoising regardless of preserving or recovering the detail information in result images. This study proposes an detail extraction network to estimate detail information from a noisy input image. Moreover, the extracted detail information is utilized to enhance the final denoised image. Experimental results demonstrate that the proposed method can outperform the existing works by a subjective measurement.

• Journal of Information Processing Systems
• /
• v.14 no.6
• /
• pp.1405-1419
• /
• 2018

An image fusion method is proposed on the basis of depth model segmentation to overcome the shortcomings of noise interference and artifacts caused by infrared and visible image fusion. Firstly, the deep Boltzmann machine is used to perform the priori learning of infrared and visible target and background contour, and the depth segmentation model of the contour is constructed. The Split Bregman iterative algorithm is employed to gain the optimal energy segmentation of infrared and visible image contours. Then, the nonsubsampled contourlet transform (NSCT) transform is taken to decompose the source image, and the corresponding rules are used to integrate the coefficients in the light of the segmented background contour. Finally, the NSCT inverse transform is used to reconstruct the fused image. The simulation results of MATLAB indicates that the proposed algorithm can obtain the fusion result of both target and background contours effectively, with a high contrast and noise suppression in subjective evaluation as well as great merits in objective quantitative indicators.

• Tunnel and Underground Space
• /
• v.34 no.3
• /
• pp.208-217
• /
• 2024

There is so much research effort for developing and implementing deep learning-based surveillance systems to manage health and safety issues in construction sites. Especially, the development of deep learning-based object detection in various environmental changes has been progressing because those affect decreasing searching performance of the model. Among the various environmental variables, the accuracy of the object detection model is significantly dropped under low illuminance, and consistent object detection accuracy cannot be secured even the model is trained using low-light images. Accordingly, there is a need of low-light enhancement to keep the performance under low illuminance. Therefore, this paper conducts a comparative study of various deep learning-based low-light image enhancement models (GLADNet, KinD, LLFlow, Zero-DCE) using the acquired construction site image data. The low-light enhanced image was visually verified, and it was quantitatively analyzed by adopting image quality evaluation metrics such as PSNR, SSIM, Delta-E. As a result of the experiment, the low-light image enhancement performance of GLADNet showed excellent results in quantitative and qualitative evaluation, and it was analyzed to be suitable as a low-light image enhancement model. If the low-light image enhancement technique is applied as an image preprocessing to the deep learning-based object detection model in the future, it is expected to secure consistent object detection performance in a low-light environment.

• The Journal of Bigdata
• /
• v.6 no.1
• /
• pp.197-206
• /
• 2021

GPR data is used for underground exploration. The data gathered are interpreted by experts based on experience as the underground facilities often reflect GPR. In addition, GPR data are different in the noise and characteristics of the data depending on the equipment, environment, etc. This often results in insufficient data with accurate labels. Generally, a large amount of training data have to be obtained to apply CNN models that exhibit high performance in image classification problems. However, due to the characteristics of GPR data, it makes difficult to obtain sufficient data. Finally, this makes neural networks unable to learn based on general supervised learning methods. This paper proposes an image classification method considering data characteristics to ensure that the accuracy of each label is similar. The proposed method is based on semi-supervised learning, and the image is classified using clustering techniques after extracting the feature values of the image from the neural network. This method can be utilized not only when the amount of the labeled data is insufficient, but also when labels that depend on the data are not highly reliable.

• Journal of the Korea Society of Computer and Information
• /
• v.27 no.9
• /
• pp.21-32
• /
• 2022

Image labeling must be preceded in order to perform object detection, and this task is considered a significant burden in building a deep learning model. Tens of thousands of images need to be trained for building a deep learning model, and human labelers have many limitations in labeling these images manually. In order to overcome these difficulties, this study proposes a method to perform object detection without significant performance degradation, even though labeling some images rather than the entire image. Specifically, in this study, low-resolution oriental painting images are converted into high-quality images using a super-resolution algorithm, and the effect of SSIM and PSNR derived in this process on the mAP of object detection is analyzed. We expect that the results of this study can contribute significantly to constructing deep learning models such as image classification, object detection, and image segmentation that require efficient image labeling.

• The Korean Journal of Nuclear Medicine Technology
• /
• v.24 no.1
• /
• pp.15-19
• /
• 2020

Purpose Generative Adversarial Network(GAN) is one of deep learning technologies. This is a way to create a real fake image after learning the real image. In this study, after acquiring artificial intelligence images through GAN, We were compared and evaluated with real scan time images. We want to see if these technologies are potentially useful. Materials and Methods 30 patients who underwent ¹⁸F-FDG Brain PET/CT scanning at Severance Hospital, were acquired in 15-minute List mode and reconstructed into 1,2,3,4,5 and 15minute images, respectively. 25 out of 30 patients were used as learning images for learning of GAN and 5 patients used as verification images for confirming the learning model. The program was implemented using the Python and Tensorflow frameworks. After learning using the Pix2Pix model of GAN technology, this learning model generated artificial intelligence images. The artificial intelligence image generated in this way were evaluated as Mean Square Error(MSE), Peak Signal to Noise Ratio(PSNR), and Structural Similarity Index(SSIM) with real scan time image. Results The trained model was evaluated with the verification image. As a result, The 15-minute image created by the 5-minute image rather than 1-minute after the start of the scan showed a smaller MSE, and the PSNR and SSIM increased. Conclusion Through this study, it was confirmed that AI imaging technology is applicable. In the future, if these artificial intelligence imaging technologies are applied to nuclear medicine imaging, it will be possible to acquire images even with a short scan time, which can be expected to reduce artifacts caused by patient movement and increase the efficiency of the scanning room.

• Electronics and Telecommunications Trends
• /
• v.33 no.4
• /
• pp.43-53
• /
• 2018

As face recognition (FR) has been well studied over the past decades, FR technology has been applied to many real-world applications such as surveillance and biometric systems. However, in the real-world scenarios, FR performances have been known to be significantly degraded owing to variations in face images, such as the pose, illumination, and low-resolution. Recently, visual intelligence technology has been rapidly growing owing to advances in deep learning, which has also improved the FR performance. Furthermore, the FR performance based on deep learning has been reported to surpass the performance level of human perception. In this article, we discuss deep-learning based high-performance FR technologies in terms of representative deep-learning based FR architectures and recent FR algorithms robust to face image variations (i.e., pose-robust FR, illumination-robust FR, and video FR). In addition, we investigate big face image datasets widely adopted for performance evaluations of the most recent deep-learning based FR algorithms.