• Proceedings of the Korea Information Processing Society Conference
• /
• 2023.11a
• /
• pp.628-631
• /
• 2023

We present a new approach to evaluate the generated texts by Large Language Models (LLMs) for meme classification. Analyzing an image with embedded texts, i.e. meme, is challenging, even for existing state-of-the-art computer vision models. By leveraging large image-to-text models, we can extract image descriptions that can be used in other tasks, such as classification. In our methodology, we first generate image captions using BLIP-2 models. Using these captions, we use GPT-4 to evaluate the relationship between the caption and the meme text. The results show that OPT_6.7B provides a better rating than other LLMs, suggesting that the proposed method has a potential for meme classification.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.10 no.4
• /
• pp.1807-1823
• /
• 2016

Local Binary Pattern (LBP) and its variants have powerful discriminative capabilities but most of them just consider each LBP code independently. In this paper, we propose sub oriented histograms of LBP for smoke detection and image classification. We first extract LBP codes from an image, compute the gradient of LBP codes, and then calculate sub oriented histograms to capture spatial relations of LBP codes. Since an LBP code is just a label without any numerical meaning, we use Hamming distance to estimate the gradient of LBP codes instead of Euclidean distance. We propose to use two coordinates systems to compute two orientations, which are quantized into discrete bins. For each pair of the two discrete orientations, we generate a sub LBP code map from the original LBP code map, and compute sub oriented histograms for all sub LBP code maps. Finally, all the sub oriented histograms are concatenated together to form a robust feature vector, which is input into SVM for training and classifying. Experiments show that our approach not only has better performance than existing methods in smoke detection, but also has good performance in texture classification.

• Journal of the Korean Society of Costume
• /
• v.61 no.5
• /
• pp.77-92
• /
• 2011

Public uniforms play an important role in creating the color and image of urban environment beyond the level of simple working clothes for unity and practicality. Hence this study aims to compare and analyze the color characteristics and images of the police officer, fire fighter, and street cleaner uniforms at home and abroad to guarantee the wearer's safety, increase convenience and agreeability, and give emotional satisfaction to citizens in harmony with urban environment color as public uniforms. For study methods, literature review investigated urban environment, color, and uniforms as public design. Empirical study extracted color data from the public worker uniforms of the world's top 20 cities selected by 'Newsweek' in 2010 and analyzed their colors, tones, and color images. The results of this study are as follows. The most common color for police uniforms were PB(dk) as the main color, PB(p) as the sub color, and Wh as the accent color. For fire fighter uniforms, PB(dkg) was the most common main color; GY(v), as the sub color; and ItGy, as the accent color. The most common color for street cleaner uniforms was YR(v) as the main color; GY(v), as the sub color; and mGy, as the accent color. As a result of analyzing color images of these uniforms, it was found that police uniforms commonly used a modern image; fire fighter uniforms commonly used a natural image and a cool casual image; and street cleaner uniforms commonly used a casual image. As examined above, the color of public uniforms suitable for urban environment and job characteristics is very important to establish the image of public institutions, as well as to create an urban image.

• Journal of Soil and Groundwater Environment
• /
• v.21 no.3
• /
• pp.35-48
• /
• 2016

A micromodel was applied to estimate the effects of geological conditions and injection methods on displacement of resident porewater by injecting scCO₂ in the pore scale. Binary images from image analysis were used to distinguish scCO₂-filled-pores from other pore structure. CO₂ flooding followed by porewater displacement, fingering migration, preferential flow and bypassing were observed during scCO₂ injection experiments. Effects of pressure, temperature, salinity, flow rate, and injection methods on storage efficiency in micromodels were represented and examined in terms of areal displacement efficiency. The measurements revealed that the areal displacement efficiency at equilibrium decreases as the salinity increases, whereas it increases as the pressure and temperature increases. It may result from that the overburden pressure and porewater salinity can affect the CO₂ solubility in water and the hydrophilicity of silica surfaces, while the neighboring temperature has a significant effect on viscosity of scCO₂. Increased flow rate could create more preferential flow paths and decrease the areal displacement efficiency. Compared to the continuous injection of scCO₂, the pulse-type injection reduced the probability for occurrence of fingering, subsequently preferential flow paths, and recorded higher areal displacement efficiency. More detailed explanation may need further studies based on closer experimental observations.

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.37 no.1
• /
• pp.55-63
• /
• 2000

This paper investigated the sub-region compression effect of the three dimensional DCT(3D-DCT) using the difference component(DC) of inter-frame in images. The proposed algorithm are the method that obtain compression effect to divide the information into subband after 3D-DCT, the data appear the type of cubic block(8${\times}$8${\times}$8) in eight difference components per unit. In the frequence domain that transform the eight differential component frames into eight DCT frames with components of both spatial and temporal frequencies of inter-frame, the image data are divided into frame component(8${\times}$8 block) of time-axis direction into 4${\times}$4 sub block in order to effectively obtain compression data because image components are concentrate in corner region with low-frequency of cubic block. Here, using the weight of sub block, we progressed compression ratio as consider to adaptive sub-region of low frequency part. In simulation, we estimated compression ratio, reconstructed image resolution(PSNR) with the simpler image and the complex image contained the higher frequency component. In the result, we could obtain the high compression effect of 30.36dB(average value in the complex-image) and 34.75dB(average value in the simple-image) in compression range of 0.04~0.05bpp.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.12 no.5
• /
• pp.205-211
• /
• 2012

By using an adaptive threshold and weight based on the wavelet transform and Haar transform, a novel image enhancement algorithm is proposed. First, a medical image was decomposed with wavelet transform and all high-frequency sub-images were decomposed with Haar transform. Secondly, noise in the frequency domain was reduced by the proposed soft-threshold method. Thirdly, high-frequency coefficients were enhanced by the proposed weight values in different sub-images. Then, the enhanced image was obtained through the inverse Haar transform and wavelet transform. But the pixel range of the enhanced image is narrower than a normal image. Lastly, the image's histogram was stretched by nonlinear histogram equalization. Experiments showed that the proposed method can be not only enhance an image's details but can also preserve its edge features effectively.

• Investigative Magnetic Resonance Imaging
• /
• v.24 no.3
• /
• pp.123-131
• /
• 2020

Purpose: Joint label fusion (JLF) is a popular multi-atlas-based segmentation algorithm, which compensates for dependent errors that may exist between atlases. However, in order to get good segmentation results, it is very important to set the several free parameters of the algorithm to optimal values. In this study, we first investigate the feasibility of a JLF algorithm for prostate segmentation in MR images, and then suggest the optimal set of parameters for the automatic prostate segmentation by validating the results of each parameter combination. Materials and Methods: We acquired T2-weighted prostate MR images from 20 normal heathy volunteers and did a series of cross validations for every set of parameters of JLF. In each case, the atlases were rigidly registered for the target image. Then, we calculated their voting weights for label fusion from each combination of JLF's parameters (r_pxy, r_pz, r_sxy, r_sz, β). We evaluated the segmentation performances by five validation metrics of the Prostate MR Image Segmentation challenge. Results: As the number of voxels participating in the voting weight calculation and the number of referenced atlases is increased, the overall segmentation performance is gradually improved. The JLF algorithm showed the best results for dice similarity coefficient, 0.8495 ± 0.0392; relative volume difference, 15.2353 ± 17.2350; absolute relative volume difference, 18.8710 ± 13.1546; 95% Hausdorff distance, 7.2366 ± 1.8502; and average boundary distance, 2.2107 ± 0.4972; in parameters of r_pxy = 10, r_pz = 1, r_sxy = 3, r_sz = 1, and β = 3. Conclusion: The evaluated results showed the feasibility of the JLF algorithm for automatic segmentation of prostate MRI. This empirical analysis of segmentation results by label fusion allows for the appropriate setting of parameters.

• Restorative Dentistry and Endodontics
• /
• v.46 no.2
• /
• pp.23.1-23.11
• /
• 2021

Objectives: This study assessed the reliability of digital color measurements using images of resin composite specimens captured with a cellphone. Materials and Methods: The reference color of cylindrical specimens built-up with the use of resin composite (shades A1, A2, A3, and A4) was measured with a portable spectrophotometer (CIELab). Images of the specimens were obtained individually or pairwise (compared shades in the same photograph) under standardized parameters. The color of the specimens was measured in the images using RGB system and converted to CIELab system using image processing software. Whiteness index (WI_D) and color differences (ΔE₀₀) were calculated for each color measurement method. For the cellphone, the ΔE₀₀ was calculated between the pairs of shades in separate images and in the same image. Data were analyzed using 2-way repeated-measures analysis of variance (α = 0.05). Linear regression models were used to predict the reference ΔE₀₀ values of those calculated using color measured in the images. Results: Images captured with the cellphone resulted in different WI_D values from the spectrophotometer only for shades A3 and A4. No difference to the reference ΔE₀₀ was observed when individual images were used. In general, a similar ranking of ΔE₀₀ among resin composite shades was observed for all methods. Stronger correlation coefficients with the reference ΔE₀₀ were observed using individual than pairwise images. Conclusions: This study showed that the use of cellphone images to measure the color difference seems to be a feasible alternative providing outcomes similar to those obtained with the spectrophotometer.

• Journal of Korea Multimedia Society
• /
• v.11 no.2
• /
• pp.153-162
• /
• 2008

We propose a novel image inpainting method composed of two parts: band matching and seamless cloning. In band matching, a band enclosing the boundary of a missing region is compared to those from the other parts of the image. The inner area of the minimum difference band is then copied to the missing region. Even though this band matching results in successful inpainting in many practical applications, brightness discontinuity (a seam) may appear between the filled missing region and its neighborhood. We apply seamless cloning to remove such discontinuity between the two regions. However, since this basic method using one patch may not deal with cases where there are abrupt changes of color or brightness along the boundary, we furthermore devise one more step: target sub-division. The target area is subdivided into small sub-areas, and the band matching and seamless cloning are applied to each of them. The multiple results from the sub-division are then ordered according to inpainting quality, which is measured based on the edge map or discontinuity map along the boundary band.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.17 no.11
• /
• pp.2645-2652
• /
• 2013

For a variety of vision systems such as car omni-directional surveillance systems and robot vision systems, many cameras have been equipped and used. In order to detect corners of grid pattern in AVM(Around View Monitoring) systems, after the non-linear radial distortion image obtained from wide-angle camera is corrected, corners of grids of the distortion corrected image must be detected. Though there are transformations such as Sub-Pixel and Hough transformation as corner detection methods for AVM systems, it is difficult to achieve automatic detection by Sub-Pixel and accuracy by Hough transformation. Therefore, we showed that the automatic detection proposed in this paper, which detects corners accurately from the distortion corrected image could be applied for AVM systems, by designing and implementing it, and evaluating its performance.