• Journal of Digital Contents Society
• /
• v.19 no.6
• /
• pp.1197-1205
• /
• 2018

In this paper, we perform Single Image Super Resolution(SISR) for acquired images of EOTS or IRST from naval combat system. In order to conduct super resolution, we use Generative Adversarial Networks(GANs), which consists of a generative model to create a super-resolution image from the given low-resolution image and a discriminative model to determine whether the generated super-resolution image is qualified as a high-resolution image by adjusting various learning parameters. The learning parameters consist of a crop size of input image, the depth of sub-pixel layer, and the types of training images. Regarding evaluation method, we apply not only general image quality metrics, but feature descriptor methods. As a result, a larger crop size, a deeper sub-pixel layer, and high-resolution training images yield good performance.

• Journal of Multimedia Information System
• /
• v.8 no.2
• /
• pp.101-110
• /
• 2021

In computer vision, single-image super resolution has been an area of research for a significant period. Traditional techniques involve interpolation-based methods such as Nearest-neighbor, Bilinear, and Bicubic for image restoration. Although implementations of convolutional neural networks have provided outstanding results in recent years, efficiency and single model multi-scalability have been its challenges. Furthermore, previous works haven't placed enough emphasis on real-number scalability. Interpolation-based techniques, however, have no limit in terms of scalability as they are able to upscale images to any desired size. In this paper, we propose a convolutional neural network possessing the advantages of the interpolation-based techniques, which is also efficient, deeming it suitable in practical implementations. It consists of convolutional layers applied on the low-resolution space, post-up-sampling along the end hidden layers, and additional layers on high-resolution space. Up-sampling is applied on a multiple channeled feature map via bicubic interpolation using a single model. Experiments on architectural structure, layer reduction, and real-number scale training are executed with results proving efficient amongst multi-scale learning (including scale multi-path-learning) based models.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.4 no.2
• /
• pp.117-137
• /
• 2010

Iris recognition is a biometric technique which uses unique iris patterns between the pupil and sclera. The advantage of iris recognition lies in high recognition accuracy; however, for good performance, it requires the diameter of the iris to be greater than 200 pixels in an input image. So, a conventional iris system uses a camera with a costly and bulky zoom lens. To overcome this problem, we propose a new method to restore a low resolution iris image into a high resolution image using a single image. This study has three novelties compared to previous works: (i) To obtain a high resolution iris image, we only use a single iris image. This can solve the problems of conventional restoration methods with multiple images, which need considerable processing time for image capturing and registration. (ii) By using bilinear interpolation and a constrained least squares (CLS) filter based on the degradation model, we obtain a high resolution iris image with high recognition performance at fast speed. (iii) We select the optimized parameters of the CLS filter and degradation model according to the zoom factor of the image in terms of recognition accuracy. Experimental results showed that the accuracy of iris recognition was enhanced using the proposed method.

• Journal of Multimedia Information System
• /
• v.8 no.4
• /
• pp.203-210
• /
• 2021

Deep leaning convolutional neural networks (CNN) have successfully been applied to image super-resolution (SR). Despite their great performances, SR techniques tend to focus on a certain upscale factor when training a particular model. Algorithms for single model multi-scale networks can easily be constructed if images are upscaled prior to input, but sub-pixel convolution upsampling works differently for each scale factor. Recent SR methods employ multi-scale and multi-path learning as a solution. However, this causes unshared parameters and unbalanced parameter distribution across various scale factors. We present a multi-scale single-path upsample module as a solution by exploiting the advantages of sub-pixel convolution and interpolation algorithms. The proposed model employs sub-pixel convolution for the highest scale factor among the learning upscale factors, and then utilize 1-dimension interpolation, compressing the learned features on the channel axis to match the desired output image size. Experiments are performed for the single-path upsample module, and compared to the multi-path upsample module. Based on the experimental results, the proposed algorithm reduces the upsample module's parameters by 24% and presents slightly to better performance compared to the previous algorithm.

• Journal of Broadcast Engineering
• /
• v.25 no.2
• /
• pp.242-251
• /
• 2020

Recently, many deep convolutional neural networks for image super-resolution have been studied. Existing deep learning-based super-resolution algorithms are architecture that up-samples the resolution at the end of the network. The post-upsampling architecture has an inefficient structure at large scaling factor result of predicting a lot of information for mapping from low-resolution to high-resolution at once. In this paper, we propose a single image super-resolution using Channel Attention Residual Dense Block based on an iterative up-down sampling architecture. The proposed algorithm efficiently predicts the mapping relationship between low-resolution and high-resolution, and shows up to 0.14dB performance improvement and enhanced subjective image quality compared to the existing algorithm at large scaling factor result.

• Nuclear Engineering and Technology
• /
• v.53 no.7
• /
• pp.2341-2347
• /
• 2021

Because single-photon emission computed tomography (SPECT) is one of the widely used nuclear medicine imaging systems, it is extremely important to acquire high-quality images for diagnosis. In this study, we designed a super-resolution (SR) technique using dense block-based deep convolutional neural network (CNN) and evaluated the algorithm on real SPECT phantom images. To acquire the phantom images, a real SPECT system using a^99mTc source and two physical phantoms was used. To confirm the image quality, the noise properties and visual quality metric evaluation parameters were calculated. The results demonstrate that our proposed method delivers a more valid SR improvement by using dense block-based deep CNNs as compared to conventional reconstruction techniques. In particular, when the proposed method was used, the quantitative performance was improved from 1.2 to 5.0 times compared to the result of using the conventional iterative reconstruction. Here, we confirmed the effects on the image quality of the resulting SR image, and our proposed technique was shown to be effective for nuclear medicine imaging.

• Journal of International Society for Simulation Surgery
• /
• v.1 no.2
• /
• pp.57-61
• /
• 2014

We present a robust 3D facial reconstruction method using a single image generated by face-specific super resolution technique. Based on the several consecutive frames with low resolution, we generate a single high resolution image and a three dimensional facial model based on it. To do this, we apply PME method to compute patch similarities for SR after two-phase warping according to facial attributes. Based on the SRI, we extract facial features automatically and reconstruct 3D facial model with basis which selected adaptively according to facial statistical data less than a few seconds. Thereby, we can provide the facial image of various points of view which cannot be given by a single point of view of a camera.

• Journal of Broadcast Engineering
• /
• v.24 no.4
• /
• pp.633-642
• /
• 2019

Recently, the convolution neural network (CNN) model at a single image super-resolution (SISR) have been very successful. The residual learning method can improve training stability and network performance in CNN. In this paper, we propose a SISR using recursive residual network architecture by introducing dense skip connections for learning nonlinear mapping from low-resolution input image to high-resolution target image. The proposed SISR method adopts a method of the recursive residual learning to mitigate the difficulty of the deep network training and remove unnecessary modules for easier to optimize in CNN layers because of the concise and compact recursive network via dense skip connection method. The proposed method not only alleviates the vanishing-gradient problem of a very deep network, but also get the outstanding performance with low complexity of neural network, which allows the neural network to perform training, thereby exhibiting improved performance of SISR method.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2018.06a
• /
• pp.63-66
• /
• 2018

Recently, a convolutional neural network (CNN) models at single image super-resolution have been very successful. Residual learning improves training stability and network performance in CNN. In this paper, we compare four convolutional neural network models for super-resolution (SR) to learn nonlinear mapping from low-resolution (LR) input image to high-resolution (HR) target image. Four models include general CNN model, global residual learning CNN model, local residual learning CNN model, and the CNN model with global and local residual learning. Experiment results show that the results are greatly affected by how skip connections are connected at the basic CNN network, and network trained with only global residual learning generates highest performance among four models at objective and subjective evaluations.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2023.05a
• /
• pp.479-481
• /
• 2023

Generative Adversarial Networks (GANs) have facilitated substantial improvement in single-image super-resolution (SR) by enabling the generation of photo-realistic images. However, the high memory requirements of GAN-based SRs (mainly generators) lead to reduced performance and increased energy consumption, making it difficult to implement them onto resource-constricted devices. In this study, we propose an efficient and compressed architecture for the SR-GAN (generator) model using the model compression technique Knowledge Distillation. Our approach involves the transmission of knowledge from a heavy network to a lightweight one, which reduces the storage requirement of the model by 58% with also an increase in their performance. Experimental results on various benchmarks indicate that our proposed compressed model enhances performance with an increase in PSNR, SSIM, and image quality respectively for x4 super-resolution tasks.