• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2013.05a
• /
• pp.385-388
• /
• 2013

In this paper, we show the weakness of the image matching method by using MPEG-7 EHD(Edge Histogram Descriptor) and suggest how to improve this weakness by using image normalization. EHD algorithm is an image matching technique that collects edge's slope of distribution and same distribution. However, the EHD error rate is high because EHD is sensitive for changes of object distortion and rotation that will be matched. We improve matching performance by accurately extract edge information in image by using normalization. We compare and analyze the normalized EHD algorithm by using distortion and rotation and it shows robustness for changes of the size and rotation.

• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.32B no.7
• /
• pp.974-988
• /
• 1995

In this paper, we propose a range image segmentation algorithm using robust regression. We derive a least $\kappa$th-order square (LKS) method by generalizing the least median of squares (LMedS) method and compare it with the conventional robust regressions. The LKS is robuster against outliers than the LMedS and shows performance similar to the residual consensus (RESC). The RESC uses the predetermined number of sorted residuals, whereas the LKS uses an adaptive parameter determined by given observations rather than the a priori knowledge. Computer simulation with synthetic and real range images shows that the proposed LKS algorithm gives better performance than the conventional ones.

• The Journal of Korea Robotics Society
• /
• v.2 no.2
• /
• pp.129-136
• /
• 2007

This paper introduces a position-based robust visual servoing method which is developed for operation of a human-like robot with two arms. The proposed visual servoing method utilizes SIFT algorithm for object detection and CAMSHIFT algorithm for object tracking. While the conventional CAMSHIFT has been used mainly for object tracking in a 2D image plane, we extend its usage for object tracking in 3D space, by combining the results of CAMSHIFT for two image plane of a stereo camera. This approach shows a robust and dependable result. Once the robot's task is defined based on the extracted 3D information, the robot is commanded to carry out the task. We conduct several position-based visual servoing tasks and compare performances under different conditions. The results show that the proposed visual tracking algorithm is simple but very effective for position-based visual servoing.

• Journal of Korea Multimedia Society
• /
• v.21 no.12
• /
• pp.1387-1395
• /
• 2018

Core algorithm of deep learning Convolutional Neural Network(CNN) shows better performance than other machine learning algorithms. However, if there is not sufficient data, CNN can not achieve satisfactory performance even if the classifier is excellent. In this situation, it has been proven that the use of transfer learning can have a great effect. In this paper, we apply two transition learning methods(freezing, retraining) to three CNN models(ResNet-50, Inception-V3, DenseNet-121) and compare and analyze how the classification performance of CNN changes according to the methods. As a result of statistical significance test using various evaluation indicators, ResNet-50, Inception-V3, and DenseNet-121 differed by 1.18 times, 1.09 times, and 1.17 times, respectively. Based on this, we concluded that the retraining method may be more effective than the freezing method in case of transition learning in image classification problem.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2019.05a
• /
• pp.558-560
• /
• 2019

Recently, there are steadily required more highly broadcasting service based on the broadcasting extension and improved quality. This paper proposed the method of the implemented above mentioned. The proposed high speed data link techniques method is more improved processing speed and more efficiently editing advantage compare with earlier methods.

• Journal of Power System Engineering
• /
• v.20 no.4
• /
• pp.32-37
• /
• 2016

This paper proposes two measurement methods for injured rate of fish surface using color image segmentation method based on K-means clustering algorithm and Otsu's threshold algorithm. To do this task, the following steps are done. Firstly, an RGB color image of the fish is obtained by the CCD color camera and then converted from RGB to HSI. Secondly, the S channel is extracted from HSI color space. Thirdly, by applying the K-means clustering algorithm to the HSI color space and applying the Otsu's threshold algorithm to the S channel of HSI color space, the binary images are obtained. Fourthly, morphological processes such as dilation and erosion, etc. are applied to the binary image. Fifthly, to count the number of pixels, the connected-component labeling is adopted and the defined injured rate is gotten by calculating the pixels on the labeled images. Finally, to compare the performances of the proposed two measurement methods based on the K-means clustering algorithm and the Otsu's threshold algorithm, the edge detection of the final binary image after morphological processing is done and matched with the gray image of the original RGB image obtained by CCD camera. The results show that the detected edge of injured part by the K-means clustering algorithm is more close to real injured edge than that by the Otsu' threshold algorithm.

• The Korean Journal of Applied Statistics
• /
• v.24 no.4
• /
• pp.695-708
• /
• 2011

Image fusion is the process of combining multiple images of the same scene into a single fused image with application to many fields, such as remote sensing, computer vision, robotics, medical imaging and military affairs. The widely used image fusion rules that use wavelet transform have been based on a simple comparison with the activity measures of local windows such as mean and standard deviation. In this case, information features from the original images are excluded in the fusion image and distorted fusion images are obtained for noisy images. In this paper, we propose the use of a nonparametric squared ranks test on the quality of variance for two samples in order to overcome the influence of the noise and guarantee the homogeneity of the fused image. We evaluate the method both quantitatively and qualitatively for image fusion as well as compare it to some existing fusion methods. Experimental results indicate that the proposed method is effective and provides satisfactory fusion results.

• The Transactions of The Korean Institute of Electrical Engineers
• /
• v.67 no.5
• /
• pp.669-674
• /
• 2018

Data augmentation is an efficient way to reduce overfitting on models and to improve a performance supplementing extra data for training. It is more important in deep learning based industrial machine vision. Because deep learning requires huge scale of learning data to learn a model, but acquisition of data can be limited in most of industrial applications. A very generic method for augmenting image data is to perform geometric transformations, such as cropping, rotating, translating and adjusting brightness of the image. The effectiveness of data augmentation in image classification has been reported, but it is rare in defect inspections. We explore and compare various basic augmenting operations for the metal surface defects. The experiments were executed for various types of defects and different CNN networks and analysed for performance improvements by the data augmentations.

• Proceedings of the KIEE Conference
• /
• 2005.10b
• /
• pp.180-182
• /
• 2005

Fingerprint-based identification is known to be used for a very long time. Owing to their uniqueness and immutability, fingerprints are today the most widely used biometric features. Therefore, recognition using fingerprints is one of the safest methods as a way of personal identification. But fingerprint identification system has a critical weakness. Since the fingerprint identification time dramatically increase when we compare the unknown fingerprint's minutiae with fingerprint database's minutiae. In this paper, a ridge orientation extraction method using Index table is proposed to solve the problem. The goal of fast direction image extraction is to reduce the identification time and to improve the clarity of ridge and valley structures of input fingerprint image.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2009.01a
• /
• pp.450-455
• /
• 2009

Amplitude-only logarithmic Radon transform (ALR transform) for pattern matching is proposed. This method provides robustness for object translation, scaling, and rotation. An ALR image is invariant even if objects are translated in a picture. For the object scaling and rotation, the ALR image is merely translated. The objects are identified using a phase-only matched filter to the ALR image. The ratio of size, the difference of rotation angle, and the position between the two objects are detected. Our pattern matching procedure is described, herein, and its simulation is executed. We compare matching scores with the Fourier-Mellin transform, and the general phase-only matched filter.