• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.22 no.5
• /
• pp.185-191
• /
• 2022

In this study, we propose a novel super-resolution neural network based on U-Net, residual neural network, and sub-pixel convolution. To prevent the loss of detailed information due to the max pooling of U-Net, we propose down-sampling and connection using sub-pixel convolution. This uses all pixels in the filter, unlike the max pooling that creates a new feature map with only the max value in the filter. As a 2×2 size filter passes, it creates a feature map consisting only of pixels in the upper left, upper right, lower left, and lower right. This makes it half the size and quadruple the number of feature maps. And we propose two methods to reduce the computation. The first uses sub-pixel convolution, which has no computation, and has better performance, instead of up-convolution. The second uses a layer that adds two feature maps instead of the connection layer of the U-Net. Experiments with a banchmark dataset show better PSNR values on all scale and benchmark datasets except for set5 data on scale 2, and well represent local area patterns.

• Journal of the Institute of Electronics and Information Engineers
• /
• v.51 no.11
• /
• pp.155-164
• /
• 2014

Local histogram equalization is one of the most popular ways of enhancing the local brightness features of an input image. However, local histogram equalization reveals some problems. First, undesired artifacts are produced by over-enhancing the local features. Second, the enhancement of local features does not always result in global contrast enhancement. To cope with these problems, we propose an illumination driven local histogram equalization method. First, to estimate the illumination information, the proposed method combines the input image and the blurred image produced through the process of the down-sampling and the up-sampling. Next, the proposed method adaptively adjusts the mapping function estimated by the local histogram equalization using the information of the illumination. As a result, the proposed illumination information driven local histogram equalization method simultaneously enhances the global and the local contrast levels while preventing any local artifacts. Experimental results show that the proposed algorithm outperforms the conventional methods on objective and subjective criteria.

• Smart Structures and Systems
• /
• v.32 no.3
• /
• pp.135-151
• /
• 2023

Using 3D asphalt pavement surface data, a deep and multiscale network named CrackNet-M is proposed in this paper for pixel-level crack detection for improvements in both accuracy and robustness. The CrackNet-M consists of four function-specific architectural modules: a central branch net (CBN), a crack map enhancement (CME) module, three pooling feature pyramids (PFP), and an output layer. The CBN maintains crack boundaries using no pooling reductions throughout all convolutional layers. The CME applies a pooling layer to enhance potential thin cracks for better continuity, consuming no data loss and attenuation when working jointly with CBN. The PFP modules implement direct down-sampling and pyramidal up-sampling with multiscale contexts specifically for the detection of thick cracks and exclusion of non-crack patterns. Finally, the output layer is optimized with a skip layer supervision technique proposed to further improve the network performance. Compared with traditional supervisions, the skip layer supervision brings about not only significant performance gains with respect to both accuracy and robustness but a faster convergence rate. CrackNet-M was trained on a total of 2,500 pixel-wise annotated 3D pavement images and finely scaled with another 200 images with full considerations on accuracy and efficiency. CrackNet-M can potentially achieve crack detection in real-time with a processing speed of 40 ms/image. The experimental results on 500 testing images demonstrate that CrackNet-M can effectively detect both thick and thin cracks from various pavement surfaces with a high level of Precision (94.28%), Recall (93.89%), and F-measure (94.04%). In addition, the proposed CrackNet-M compares favorably to other well-developed networks with respect to the detection of thin cracks as well as the removal of shoulder drop-offs.

• Journal of Broadcast Engineering
• /
• v.14 no.5
• /
• pp.638-649
• /
• 2009

Image resizing (or scaling) is one of the most essential issues for the success of visual service because image data has to be adapted to the variety of display features. For 2D imaging, the image scaling is generally accomplished by 2D image re-sampling (i.e., up-/down-sampling). However, when it comes to stereoscopic 3D images, 2D re-sampling methods are inadequate because additional consideration on the third dimension of depth is not incorporated. Practically, stereoscopic 3D image scaling is process with left/right images, not stereoscopic 3D image itself, because the left/right Images are only tangible data. In this paper, we analyze stereoscopic 3D image scaling from two aspects: geometrical deformation and frequency-domain aliasing. A number of 3D displays are available in the market and they have various screen dimensions. As we have more varieties of the displays, efficient stereoscopic 3D image scaling is becoming more emphasized. We present the recommendations for the 3D scaling from the geometric analysis and propose a disparity-adaptive filter for anti-aliasing which could occur during the image scaling process.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.10 no.8
• /
• pp.3791-3805
• /
• 2016

The visual attention mechanism includes 2 attention models, the bottom-up (B-U) and the top-down (T-D), the physiology of which have not yet been accurately described. In this paper, the visual attention mechanism is regarded as a Bayesian fusion process, and a visual attention model based on particle filter is proposed. Under certain particular assumed conditions, a calculation formula of Bayesian posterior probability is deduced. The visual attention fusion process based on the particle filter is realized through importance sampling, particle weight updating, and resampling, and visual attention is finally determined by the particle distribution state. The test results of multigroup images show that the calculation result of this model has better subjective and objective effects than that of other models.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.45 no.5
• /
• pp.52-62
• /
• 2008

In this paper, we develop a method that changes the resolutions of images in an arbitrary block transform domain by using a filter that suits to the characteristics of the underlying images. To accomplish this, we represent each procedure resizing images in an arbitrary transform domain as matrix multiplications and we derive the matrix that scales the image resolutions from the matrix multiplications. The resolution scaling matrix is also designed to be able to select the up/down-sampling filter that suits the characteristics of the image. Experiments show that the proposed method produces the reliable performances when it is applied to various transforms and to images that are mixed with predicted and non-predicted blocks which are generated during video coding.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.17 no.8
• /
• pp.1777-1783
• /
• 2013

We do research on the simulation of Doppler effect from a target's moving under the sea by Transmission Line Matrix modeling which is one of numerical methods on time domain. To implement the effect, the input signal was entered at a moving node according to a moving target's moving speed. The result had maximum 2.47% error compared with the theoretical value. And from simulation results with speed control of a moving target, we could also obtain resonable results within 0.63% error range.

• Journal of Fashion Business
• /
• v.2 no.4
• /
• pp.102-111
• /
• 1998

The purpose of this study is to present the basic material for the competitive, high-quality products by analyzing employees' attitude toward their job and some factors which influence the enhancement of productivity, education of workers, and facilities. This project was mainly conducted the conditions of women's wear manufacturing industries in Taegu by interviewing 143 employees in the sewing department. The result of this dissertation can be summarized as follows; 1. When it comes to the division of task in the sewing department, one team is composed of chairpersons (33.5%), assistant members (27.4%) and sub-assistant members (26.6%). The inspection of commodities as well as the enhancement of work on ironing for elaboration was divided as completion (12.6%) and finally the forms of task were made up of design sampling team (49%) and contracting team (51%). 2. Among the needle workers, as many as 60% employees had received technical education, which was by means of being passed down from the predecessors (81.6%). In terms of the period of education, from 2 to 5 years topped the list. The contents of education comprise patterning, sewing (65.2%) and ironing (20.3%). 3. The department of design takes charge of the report on working directions up to 88.2% and the working directions were used by 69.9 % of technicians. The directors of working conditions and methods were chiefs of designing department (37.7%) and those of sewing department (30.8%). The factors of defective goods were low-quality materials (50.0%), the deficiency in skills of workers. In terms of methods for preventing defects, technical education of workers, standardization of task and investment for factory automation were suggested. 4. As for their perception of present work, most of employees (80%) look upon it as a way of earning a living and 11% of them wanted to derive many things from their work. 59.2% of workers were satisfied with the product. When it comes to the incentive system provided by company, 67.0% responsed that it was not bad, and 23.9% evaluated it as 'satisfactory'.

• Journal of Environmental Science International
• /
• v.16 no.7
• /
• pp.855-863
• /
• 2007

In order to understand the temporal distribution of pico- and nanoplankton and factors controlling its distribution at a station in Okkye Bay of Masan Bay located in the southern part of Korea, this study was conducted on two weeks interval from April 2005 to April 2006, and several abiotic and biotic factors were measured. During the study, picoplankton consisted of picoflagellates, cyanobacteria and heterotrophic bacteria, and nanoplankton consisted of nanoflagellates excluding dinoflagellates. The concentration of chlorophyll-a (chl-a) was a mean of $4.33\;{\mu}g/L$, and the nanoplanktonic ($<20\;{\mu}m$) chl-a size fraction was a mean of 39.5 % and significantly correlated with water temperature. The abundances of cyanobacteria and photosynthetic flagellates (PF) were means of $24.4{\times}10^{3}\;cells/mL\;and\;2.87{\times}10^{3}\;cells/mL$, respectively. The contribution of picoflagellates to the PF abundance varied among the sampling occasions and was a mean of 29 %, but to the PF carbon biomass was 2.6 % only. The PF abundance had significant relationships with water temperature, and silicate and TIN concentrations, suggesting that the PF abundance seemed to be primarily bottom-up regulated. The abundance of heterotrophic bacteria was a mean of $3.18{\times}10^{6}\;cells/mL$ and unlike other ecosystems it did not have relationships with chl-a and heterotrophic flagellates (HF), suggesting that bacterial abundance did not seem to be bottom-up or top-down regulated. HF mostly consisted of cells less than $5{\mu}m$ and its abundance was a mean of $2.71{\times}10^{3}\;cells/mL$. Of the HF abundance, picoflagellates occupied about 31 %, and occupied about 9 % of the HF carbon biomass. HF grazing activity on heterotrophic bacteria was relatively low and removed about 10 % of bacterial abundance, suggesting that HF might not be major consumers of bacteria and there seems to be other consumers in Okkye Bay. These results suggest that Okkye Bay may have a unique microbial ecosystem.

• Journal of the Institute of Electronics Engineers of Korea SD
• /
• v.43 no.11 s.353
• /
• pp.37-47
• /
• 2006

This work proposes a 10b 25MS/s $0.8mm^2$ 4.8mW 0.13um CMOS A/D Converter (ADC) for high-performance wireless communication systems such as DVB, DAB and DMB simultaneously requiring low voltage, low power, and small area. A two-stage pipeline architecture minimizes the overall chip area and power dissipation of the proposed ADC at the target resolution and sampling rate while switched-bias power reduction techniques reduce the power consumption of analog amplifiers. A low-power sample-and-hold amplifier maintains 10b resolution for input frequencies up to 60MHz based on a single-stage amplifier and nominal CMOS sampling switches using low threshold-voltage transistors. A signal insensitive 3-D fully symmetric layout reduces the capacitor and device mismatch of a multiplying D/A converter while low-noise reference currents and voltages are implemented on chip with optional off-chip voltage references. The employed down-sampling clock signal selects the sampling rate of 25MS/s or 10MS/s with a reduced power depending on applications. The prototype ADC in a 0.13um 1P8M CMOS technology demonstrates the measured DNL and INL within 0.42LSB and 0.91LSB and shows a maximum SNDR and SFDR of 56dB and 65dB at all sampling frequencies up to 2SMS/s, respectively. The ADC with an active die area if $0.8mm^2$ consumes 4.8mW at 25MS/s and 2.4mW at 10MS/s at a 1.2V supply.