• Journal of the Optical Society of Korea
• /
• v.13 no.1
• /
• pp.116-122
• /
• 2009

In this paper, a robust method is developed to locate the irises of both eyes. The method doesn't put any restrictions on the background. The method is based on the AdaBoost algorithm for face and eye candidate points detection. Candidate points are tuned such that two candidate points are exactly in the centers of the irises. Mean crossing function and convolution template are proposed to filter out candidate points and select the iris pair. The advantage of using this kind of hybrid method is that AdaBoost is robust to different illumination conditions and backgrounds. The tuning step improves the precision of iris localization while the convolution filter and mean crossing function reliably filter out candidate points and select the iris pair. The proposed structure is evaluated on three public databases, Bern, Yale and BioID. Extensive experimental results verified the robustness and accuracy of the proposed method. Using the Bern database, the performance of the proposed algorithm is also compared with some of the existing methods.

• Journal of ILASS-Korea
• /
• v.10 no.2
• /
• pp.18-31
• /
• 2005

This study was performed to explore the liquid breakup and atomization characteristics for the classification of drop formation mode and background of uniform droplets generation in electrohydrodynmaic atomization according to the change of experimental parameters such as nozzle material (stainless steel. teflon). fluid flow rate, applied electrical field and intensity, and frequency. In results, from the classification map of drop formation modes according to the variation of applied AC voltage and frequency at a stainless nozzle, the droplet size was smaller than the outer diameter of the nozzle tip relatively in the spindle mode. The transition points became clearly to be moved toward the high applied voltage by rising the applied AC frequency beyond 450Hz. Also the droplet radius can be observed quite small in the frequency bandwidth of $350{\sim}450Hz$. The droplet radiuses decrease as the applied voltage increases for a fixed applied AC frequency within the range from 50Hz to 400Hz Over 400Hz, the relation between the power intensity and the droplet size was not consistent with a continuous mechanism of liquid breakup. Thus, it is showed that the droplet size distribution using the teflon nozzle was analogous to the results of stainless steel, but the droplet size was bigger than that of stainless steel relatively in case of a teflon nozzle.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.15 no.10
• /
• pp.3668-3684
• /
• 2021

Video action recognition is widely used in video surveillance, behavior detection, human-computer interaction, medically assisted diagnosis and motion analysis. However, video action recognition can be disturbed by many factors, such as background, illumination and so on. Two-stream convolutional neural network uses the video spatial and temporal models to train separately, and performs fusion at the output end. The multi segment Two-Stream convolutional neural network model trains temporal and spatial information from the video to extract their feature and fuse them, then determine the category of video action. Google Xception model and the transfer learning is adopted in this paper, and the Xception model which trained on ImageNet is used as the initial weight. It greatly overcomes the problem of model underfitting caused by insufficient video behavior dataset, and it can effectively reduce the influence of various factors in the video. This way also greatly improves the accuracy and reduces the training time. What's more, to make up for the shortage of dataset, the kinetics400 dataset was used for pre-training, which greatly improved the accuracy of the model. In this applied research, through continuous efforts, the expected goal is basically achieved, and according to the study and research, the design of the original dual-flow model is improved.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.13 no.6
• /
• pp.3182-3198
• /
• 2019

Vision-based Human Pose Estimation has been considered as one of challenging research subjects due to problems including confounding background clutter, diversity of human appearances and illumination changes in scenes. To tackle these problems, we propose to use a new multi-stage convolution machine for estimating human pose. To provide better heatmap prediction of body joints, the proposed machine repeatedly produces multiple predictions according to stages with receptive field large enough for learning the long-range spatial relationship. And stages are composed of various modules according to their strategic purposes. Pyramid stacking module and dilation module are used to handle problem of human pose at multiple scales. Their multi-scale information from different receptive fields are fused with concatenation, which can catch more contextual information from different features. And spatial and channel information of a given input are converted to gating factors by squeezing the feature maps to a single numeric value based on its importance in order to give each of the network channels different weights. Compared with other ConvNet-based architectures, we demonstrated that our proposed architecture achieved higher accuracy on experiments using standard benchmarks of LSP and MPII pose datasets.

• Science of Emotion and Sensibility
• /
• v.16 no.2
• /
• pp.195-206
• /
• 2013

The main causes of traffic accidents while driving a car is of the driver's visual distraction. In this study, the color sensitivity of the information projected on the windshield were evaluated for HUD (Head Up Display) system which helps the driver's eyes on the road while driving. The driving Information were projected $9^{\circ}$ downward from front sight $0^{\circ}$ under lab's fluorescent lights, LED floorlights and the TV had having 25 [lux] illumination when driving at night environment and 100,000 [lux] of daylight environment. Munsell color hue of the basic five colors (R, Y, G, B, P) and the color of traffic lights YR, W were the color of the seven characters, each character were outlined by White, Gray except for W. Total of 19 experimental stimuli was shown in the environment of day and night driving for asking visibility information of color, fatigue, preferences, and evaluate the degree of interference. The results came out that the bright Y and G color is visibility significantly for daylight. Second, with the outline of the text, the color of the outline works as a background for luminance contrast effects and affects visibility. Third, without the outline, the glass in front of the vehicle acts as the background and the luminance contrast of characters achieve greater brightness and visibility. The luminance contrast between the stimuli and background should be considered for increasing color visibility for driving information which is an important factor for HUD commercialization.

• Journal of Internet Computing and Services
• /
• v.16 no.1
• /
• pp.29-37
• /
• 2015

In this paper, we propose a robust pre-processing module for camera-based Optical Music Score Recognition (OMR) on mobile device. The captured images likely suffer for recognition from many distortions such as illumination, blur, low resolution, etc. Especially, the complex background music sheets recognition are difficult. Through any symbol recognition system, the staff line height and staff line space are used many times and have a big impact on recognition module. A robust and accurate staff line height and staff line space are essential. Some staff line height and staff line space are proposed for binary image. But in case of complex background music sheet image, the binarization results from common binarization algorithm are not satisfactory. It can cause incorrect staff line height and staff line space estimation. We propose a robust staff line height and staff line space estimation by using run-length encoding technique on edge image. Proposed method is composed of two steps, first step, we conducted the staff line height and staff line space estimation based on edge image using by Sobel operator on image blocks. Each column of edge image is encoded by run-length encoding algorithm Second step, we detect the staff line using by Stable Path algorithm and removal the staff line using by adaptive Line Track Height algorithm which is to track the staff lines positions. The result has shown that robust and accurate estimation is possible even in complex background cases.

• Journal of the Korean Institute of Illuminating and Electrical Installation Engineers
• /
• v.20 no.1
• /
• pp.27-33
• /
• 2006

Evaluating comfort of illumination environment of building interior is recognizing the degree of glare causing discomfort. Currently, to use the experimental formula for discomfort glare studied abroad it would be not appropriate because each races feel about the degree of glare differently. Therefore, this study aim to make up prediction formula for evaluating discomfort glare reasonably from Koreans' vision and it proceeded with 4 stages as follows: First, after reviewing the existing discomfort glare evaluation formula, I selected experimental variables. Second, I made a mock-up that I can control experimental variables and conditions according to the purpose of this study. Third, 1 conducted discomfort glare evaluation experiment. Finally, compared with UGR evaluation method suggested for Westerner in prior studies. In conclusion, 1) it's proved that discomfort glare is influenced highly by a light source luminance, background luminance and location of testee and the line of vision. 2) In interior discomfort glare experiment whether the glare light source is placed within range of vision or not has more significant influence than the distance between the light source and testee. 3) I compared and analyzed with UGR, the most representative discomfort glare evaluation system and I found there is a little difference in the results. This shows discomfort glare of Koreans and Westerners are different.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.6 no.1
• /
• pp.124-132
• /
• 2002

In this paper, we present a vision algorithm and method for input image improvement and preprocessing of dented and raised characters on the sidewall of tires. we define optical condition between reflect coefficient and reflectance by the physical vector calculate. On the contrary this work will recognize the engraved characters using the computer vision technique. Tire input images have all most same grey levels between the characters and backgrounds. The reflectance is little from a tire surface. therefore, it's very difficult segment the characters from the background. Moreover, one side of the character string is raised and the other is dented. So, the captured images are varied with the angle of camera and illumination. For optimum Input images, the angle between camera and illumination was found out to be with in 90$^{\circ}$. In addition, We used complex filtering with low-pass and high-pass band filters to improve input images, for clear input images. Finally we define equation reflect coefficient and reflectance. By doing this, we obtained good images of tires for pattern recognition.

• The Transactions of the Korea Information Processing Society
• /
• v.4 no.4
• /
• pp.1132-1140
• /
• 1997

In many vision tasks of the major obstacles is the specular highkight of smoth objects, which causes a misinterpretation of objects.This paper presents an dffcient algorithm for highight detection and object reconstruction, blsed on the theory of photometric stereo in which the location of highilight changes as the position of illumination source changes.Two images, referred to as base image and reference image.are sequentially taken with two different positionhs of the two images.The difference image is thresholded to detct the specular spike of the highlight.Then the specu-lar lobe around the specular spike is detected to reconstruct the object.The proposed algorithm can be applied to metals and dielectrics, regardlless of the surface chracteristics.This method can also be aplied to the case when the background is brighter than the object.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.8 no.8
• /
• pp.2895-2912
• /
• 2014

Content-based image retrieval has been the most important technique for managing huge amount of images. The fundamental yet highly challenging problem in this field is how to measure the content-level similarity based on the low-level image features. The primary difficulties lie in the great variance within images, e.g. background, illumination, viewpoint and pose. Intuitively, an ideal similarity measure should be able to adapt the data distribution, discover and highlight the content-level information, and be robust to those variances. Motivated by these observations, we in this paper propose a probabilistic similarity learning approach. We first model the distribution of low-level image features and derive the free energy kernel (FEK), i.e., similarity measure, based on the distribution. Then, we propose a learning approach for the derived kernel, under the criterion that the kernel outputs high similarity for those images sharing the same class labels and output low similarity for those without the same label. The advantages of the proposed approach, in comparison with previous approaches, are threefold. (1) With the ability inherited from probabilistic models, the similarity measure can well adapt to data distribution. (2) Benefitting from the content-level hidden variables within the probabilistic models, the similarity measure is able to capture content-level cues. (3) It fully exploits class label in the supervised learning procedure. The proposed approach is extensively evaluated on two well-known databases. It achieves highly competitive performance on most experiments, which validates its advantages.