• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.24 no.5
• /
• pp.674-677
• /
• 2020

The cultural background of users utilizing image search engines has a significant impact on the satisfaction of the search results. Therefore, it is important to analyze and understand the cultural context of images for more accurate image search. In this paper, we investigate how the cultural context of images can affect the performance of image classification. To this end, we first collected various types of images (e.g,. food, temple, etc.) with various cultural contexts (e.g., Korea, Japan, etc.) from web search engines. Afterwards, a deep transfer learning approach using VGG19 and MobileNetV2 pre-trained with ImageNet was adopted to learn the cultural features of the collected images. Through various experiments we show the performance of image classification can be differently affected according to the cultural context of images.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2022.11a
• /
• pp.112-115
• /
• 2022

The process of deep learning usually needs to deal with massive data which has greatly limited the development of deep learning technologies today. Convolutional Neural Network (CNN) structure is often used to solve image classification problems. However, a large number of images may be required in order to train an image in CNN, which is a heavy burden for existing computer systems to handle. If the image data can be compressed under the premise that the computer hardware system remains unchanged, it is possible to train more datasets in deep learning. However, image compression usually adopts the form of lossy compression, which will lose part of the image information. If the lost information is key information, it may affect learning performance. In this paper, we will analyze the effect of image compression on deep learning performance on CNN-based cat and dog classification. Through the experiment results, we conclude that the compression of images does not have a significant impact on the accuracy of deep learning.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2009.01a
• /
• pp.80-84
• /
• 2009

This paper investigates the relationship between the compression ratio and the gamut area for a reconstructed image when using JPEG and JPEG2000. Eighteen color samples from the Macbeth ColorChecker are initially used to analyze the relationship between the compression ratio and the color bleeding phenomenon, i.e. the hue and chroma shifts in the a*b* color plane. In addition, twelve natural color images, divided into two groups depending on four color attributes, are also used to investigate the relationship between the compression ratio and the variation in the gamut area. For each image group, the gamut area for the reconstructed image shows an overall tendency to increase when increasing the compression ratio, similar to the experimental results with the Macbeth ColorChecker samples. However, with a high compression ratio, the gamut area decreases due to the mixture of adjacent colors, resulting in more grey.

• 한국정보디스플레이학회:학술대회논문집
• /
• 2009.10a
• /
• pp.1228-1231
• /
• 2009

This paper presented an automatic system for display quality measurement by image processing. The goal is to replace human eyes for display quality evaluation by computer vision and get the objective quality review for consumer to make purchase of monitor or TV. Color, contrast, brightness, sharpness and motion blur are the main five factors to affect display quality that could be measured by supplying patterns and analyzing the corresponding images captured from webcam. The scores are calculated by image processing techniques. Linear regression model is then adopted to find the relation between human score and the measured display performance.

• Proceedings of the KIEE Conference
• /
• 1995.07b
• /
• pp.974-976
• /
• 1995

Segmentation of image data is an important problem in computer vision, remote sensing, and image analysis. Most objects in the real world have textured surfaces. Segmentation based on texture information is possible even if there are no apparent intensity edges between the different regions. There are many existing methods for texture segmentation and classification, based on different types of statistics that can be obtained from the gray-level images. In this paper, we use a neural network model --- ART-2 (Adaptive Resonance Theory) for textures in an image, proposed by Carpenter and Grossberg. In our experiments, we use Walsh matrix as feature value for textured image.

• ETRI Journal
• /
• v.24 no.6
• /
• pp.443-454
• /
• 2002

This paper introduces a system for mosaicking sequences of soccer images in a panoramic view for soccer game analysis. The continuous mosaic images of the soccer ground field allow the user to view a wide picture of the players' actions. The initial component of our algorithm automatically detects and traces the players and some lines. The next component of our algorithm finds the parameters of the captured image coordinates and transforms them into ground model coordinates for automatic soccer game analysis. The results of our experimentations indicate that the proposed system offers a promising method for segmenting, mosaicking, and analyzing soccer image sequences.

• Journal of the Korean Society for Precision Engineering
• /
• v.20 no.4
• /
• pp.136-142
• /
• 2003

The morphological analysis of wear debris can provide early a failure diagnosis in lubricated moving system. It can be effective to analyze operating conditions of oil-lubricated tribological system with shape characteristics of wear debris in a lubricant. But, in order to predict and recognize an operating condition of lubricated machine, it is needed to analyze and to identify shape characteristics of wear debris. Therefore, If the morphological characteristics of wear debris are recognized by computer image analysis using the neural network algorithm, it is possible to recognize operating condition of hydraulic driving members. In this study, wear debris in the lubricating oil are extracted by membrane filter (0.45 ${\mu}{\textrm}{m}$), and the quantitative values of shape parameters of wear debris are calculated by the digital image processing. This shape parameters are studied and identified by the artificial neural network algorithm. The result of study could be applied to prediction and to recognition of the operating condition of hydraulic driving members in lubricated machine systems.

• Journal of information and communication convergence engineering
• /
• v.13 no.4
• /
• pp.270-274
• /
• 2015

In this paper, we propose an improved framework to analyze an N-ocular imaging system under fixed constrained resources such as the number of image sensors, the pixel size of image sensors, the distance between adjacent image sensors, the focal length of image sensors, and field of view of image sensors. This proposed framework takes into consideration, for the first time, the defocusing effect of the imaging lenses according to the object distance. Based on the proposed framework, the N-ocular imaging system such as integral imaging is analyzed in terms of depth resolution using two-point-source resolution analysis. By taking into consideration the defocusing effect of the imaging lenses using ray projection model, it is shown that an improved depth resolution can be obtained near the central depth plane as the number of cameras increases. To validate the proposed framework, Monte Carlo simulations are carried out and the results are analyzed.

• Journal of the Korea Society of Computer and Information
• /
• v.27 no.1
• /
• pp.77-82
• /
• 2022

The research study was conducted for development of the advanced image analysis service system based on deep learning. CNN(Convolutional Neural Network) is built in this system to extract learning data collected from Google and Instagram. The service gets a place image of Jeju as an input and provides relevant location information of it based on its own learning data. Accuracy improvement plans are applied throughout this study. In conclusion, the implemented system shows about 79.2 of prediction accuracy. When the system has plenty of learning data, it is expected to predict various places more accurately.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.11 no.3
• /
• pp.1761-1784
• /
• 2017

In order to improve the embedding rate of reversible watermarking algorithm for digital image and enhance the imperceptibility of the watermarked image, an adaptive reversible image watermarking algorithm based on DE is proposed. By analyzing the traditional DE algorithm and the generalized DE algorithm, an improved difference expansion algorithm is proposed. Through the analysis of image texture features, the improved algorithm is used for embedding and extracting the watermark. At the same time, in order to improve the embedding capacity and visual quality, the improved algorithm is optimized in this paper. Simulation results show that the proposed algorithm can not only achieve the blind extraction, but also significantly heighten the embedded capacity and non-perception. Moreover, compared with similar algorithms, it is easy to implement, and the quality of the watermarked images is high.