• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.34 no.12C
• /
• pp.1154-1166
• /
• 2009

In this paper, we propose algorithms for generating high quality virtual intermediate views on the baseline or out of baseline. In this proposed algorithm, depth information as well as 3D warping technique is used to generate the virtual views. The coordinate of real 3D image is calculated from the depth information and geometrical characteristics of camera and the calculated 3D coordinate is projected to the 2D plane at arbitrary camera position and results in 2D virtual view image. Through the experiments, we could show that the generated virtual view image on the baseline by the proposed algorithm has better PSNR at least by 0.5dB and we also could cover the occluded regions more efficiently for the generated virtual view image out of baseline by the proposed algorithm.

• The KIPS Transactions:PartB
• /
• v.10B no.7
• /
• pp.803-810
• /
• 2003

In this paper, we describe a fast image mosaic method for constructing a large-scale image with video image captured from cameras that are arranged in radial shape. In the first step, we adopt the phase correlation algorithm to estimate the horizontal and vertical displacement between two adjacent images. Secondly, we calculate the accurate transform matrix among those cameras with Levenberg-Marquardt method. In the last step, those images are stitched into one large scale image in real-time by applying the transform matrix to the texture mapping function of DirectX. The feature of the method is that we do not need to use special hardware devices or write machine-level programs for Implementing a real-time mosaic system since we use conventional graphic APIs (Application Programming Interfaces), DirectX for image synthesis process.

• Korean Journal of Remote Sensing
• /
• v.31 no.2
• /
• pp.161-170
• /
• 2015

This paper presents an efficient image fusion method to be appropriate for the KOMPSAT-2 and 3 satellites. The proposed method is based on the well-established component substitution (CS) approach. The proposed method is divided into two parts: 1) The first step is to create a intensity image by the weighted-averaging operation of a multi-spectral (MS) image and 2) the second step is to produce an optimal high-frequency image using the statistical properties of the original MS and panchromatic (PAN) images. The performance of the proposed method is evaluated in both quantitative and visual analysis. Quantitative assessments are performed by using the relative global dimensional synthesis error (Spatial and Spectral ERGAS), the image quality index (Q4), and the spectral angle mapper index (SAM). The qualitative and quantitative assessment results show that the fusion performance of the proposed method is improved in both the spectral and spatial qualities when it is compared with previous CS-based fusion methods.

• KIPS Transactions on Software and Data Engineering
• /
• v.8 no.10
• /
• pp.403-410
• /
• 2019

Image de-snowing aims at eliminating the negative influence by snow particles and improving scene understanding in images. In this paper, a perceptual generative adversarial network based a single image snow removal method is proposed. The residual U-Net is designed as a generator to generate the snow free image. In order to handle various sizes of snow particles, the inception module with different filter kernels is adopted to extract multiple resolution features of the input snow image. Except the adversarial loss, the perceptual loss and total variation loss are employed to improve the quality of the resulted image. Experimental results indicate that our method can obtain excellent performance both on synthetic and realistic snow images in terms of visual observation and commonly used visual quality indices.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.23 no.8
• /
• pp.896-902
• /
• 2019

In this paper, we proposed two methods to automatically generate color images similar to existing images using genetic algorithms. Experiments were performed on two different sizes($256{\times}256$, $512{\times}512$) of gray and color images using each of the proposed methods. Experimental results show that there are significant differences in the evolutionary performance of each technique in genetic modeling for image generation. In the results, evolving the whole image into sub-images evolves much more effective than modeling and evolving it into a single gene, and the generated images are much more sophisticated. Therefore, we could find that gene modeling, selection method, crossover method and mutation rate, should be carefully decided in order to generate an image similar to the existing image in the future, or to learn quickly and naturally to generate an image synthesized from different images.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.26 no.8
• /
• pp.1136-1141
• /
• 2022

Researches of face recognition on masked faces have been increasingly important due to the COVID-19 pandemic. To realize a stable and practical recognition performance, large amount of facial image data should be acquired for the purpose of training. However, it is difficult for the researchers to obtain masked face images for each human subject. This paper proposes a novel method to synthesize a face image and a virtual mask pattern. In this method, a pair of masked face image and unmasked face image, that are from a single human subject, is fed into a convolutional autoencoder as training data. This allows learning the geometric relationship between face and mask. In the inference step, for a unseen face image, the learned convolutional autoencoder generates a synthetic face image with a mask pattern. The proposed method is able to rapidly generate realistic masked face images. Also, it could be practical when compared to methods which rely on facial feature point detection.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.26 no.12
• /
• pp.1919-1925
• /
• 2022

In this paper, we propose an image-to-image translation approach based on the conditional generative adversarial network for semantic segmentation. Semantic segmentation is the task of clustering parts of an image together which belong to the same object class. Unlike the traditional pixel-wise classification approach, the proposed method parses an input RGB image to its corresponding semantic segmentation mask using a pixel regression approach. The proposed method is based on the Pix2Pix image synthesis method. We employ residual connections-based convolutional neural network architectures for both the generator and discriminator architectures, as the residual connections speed up the training process and generate more accurate results. The proposed method has been trained and tested on the NYU-depthV2 dataset and could achieve a good mIOU value (49.5%). We also compare the proposed approach to the current methods in semantic segmentation showing that the proposed method outperforms most of those methods.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.29 no.1C
• /
• pp.102-110
• /
• 2004

In this paper, we propose a VLSI structure of real-time image compression and reconstruction processor using 2-D discrete wavelet transform and implement into a hardware which use minimal hardware resource using ASIC library. In the implemented hardware, Data path part consists of the DWT kernel for the wavelet transform and inverse transform, quantizer/dequantizer, the huffman encoder/huffman decoder, the adder/buffer for the inverse wavelet transform, and the interface modules for input/output. Control part consists of the programming register, the controller which decodes the instructions and generates the control signals, and the status register for indicating the internal state into the external of circuit. According to the programming condition, the designed circuit has the various selective output formats which are wavelet coefficient, quantization coefficient or index, and Huffman code in image compression mode, and Huffman decoding result, reconstructed quantization coefficient, and reconstructed wavelet coefficient in image reconstructed mode. The programming register has 16 stages and one instruction can be used for a horizontal(or vertical) filtering in a level. Since each register automatically operated in the right order, 4-level discrete wavelet transform can be executed by a programming. We synthesized the designed circuit with synthesis library of Hynix 0.35um CMOS fabrication using the synthesis tool, Synopsys and extracted the gate-level netlist. From the netlist, timing information was extracted using Vela tool. We executed the timing simulation with the extracted netlist and timing information using NC-Verilog tool. Also PNR and layout process was executed using Apollo tool. The Implemented hardware has about 50,000 gate sizes and stably operates in 80MHz clock frequency.

• Journal of Korea Society of Industrial Information Systems
• /
• v.24 no.6
• /
• pp.25-34
• /
• 2019

VTON (Virtual try-on) is a key technology that can activate the online commerce of fashion items. However, the early 3D graphics-based methods require the 3D information of the clothing or the human body, which is difficult to secure realistically. In order to overcome this problem, Image-based deep-learning algorithms such as VITON (Virtual image try-on) and CP-VTON (Characteristic preserving-virtual try-on) has been published, but only a sampled results on performance is presented. In order to examine the strength and weakness for their commercialization, the performance analysis is needed according to the complexity of the clothes, the object posture and body shape, and the degree of occlusion of the clothes. In this paper, IoU and SSIM were evaluated for the performance of transformation and synthesis stages, together with non-DL SCM based method. As a result, CP-VTON shows the best performance, but its performance varies significantly according to posture and complexity of clothes. The reasons for this were attributed to the limitations of secondary geometric deformation and the limitations of the synthesis technology through GAN.

• Journal of the Institute of Electronics and Information Engineers
• /
• v.52 no.4
• /
• pp.175-183
• /
• 2015

The proposed algorithm consists of two stages: the learning and synthesis stages. At the learning stage, we first collect various high-resolution (HR)-low-resolution (LR) text image pairs, and quantize the LR images, and extract HR-LR block pairs. Based on quantized LR blocks, the LR-HR block pairs are clustered into a pre-determined number of classes. For each class, an optimal 2D-FIR filter is computed, and it is stored into a dictionary with the corresponding LR block for indexing. At the synthesis stage, each quantized LR block in an input LR image is compared with every LR block in the dictionary, and the FIR filter of the best-matched LR block is selected. Finally, a HR block is synthesized with the chosen filter, and a final HR image is produced. Also, in order to cope with noisy environment, we generate multiple dictionaries according to noise level at the learning stage. So, the dictionary corresponding to the noise level of the input image is chosen, and a final HR image is produced using the selected dictionary. Experimental results show that the proposed algorithm outperforms the previous works for noisy images as well as noise-free images.