• Journal of the Institute of Convergence Signal Processing
• /
• v.14 no.1
• /
• pp.33-38
• /
• 2013

In this paper, we propose an algorithm to track multi-small targets in infrared image sequences in case of dissipation or creation of targets by using the background estimation filter, Kahnan filter and mean shift algorithm. We detect target candidates in a still image by subtracting an original image from an background estimation image, and we track multi-targets by using Kahnan filter and target selection. At last, we adjust specific position of targets by using mean shift algorithm In the experiments, we compare the performance of each background estimation filters, and verified that proposed algorithm exhibits better performance compared to classic methods.

• Smart Media Journal
• /
• v.8 no.2
• /
• pp.46-57
• /
• 2019

In recent research, deep learning-based methods have achieved state-of-the-art performance in various computer vision tasks. However, these methods are commonly supervised, and require huge amounts of annotated data to train. Acquisition of data demands an additional costly effort, particularly for the tasks where it becomes challenging to obtain large amounts of data considering the time constraints and the requirement of professional human diligence. In this paper, we present a data level synthetic sampling solution to learn from small and imbalanced data sets using Generative Adversarial Networks (GANs). The reason for using GANs are the challenges posed in various fields to manage with the small datasets and fluctuating amounts of samples per class. As a result, we present an approach that can improve learning with respect to data distributions, reducing the partiality introduced by class imbalance and hence shifting the classification decision boundary towards more accurate results. Our novel method is demonstrated on a small dataset of 2789 tomato plant disease images, highly corrupted with class imbalance in 9 disease categories. Moreover, we evaluate our results in terms of different metrics and compare the quality of these results for distinct classes.

• Journal of Korea Multimedia Society
• /
• v.10 no.12
• /
• pp.1559-1565
• /
• 2007

The many target-detection methods that use forward-looking infrared (FUR) images can deal with large targets measuring $70{\times}40$ pixels, utilizing their shape features. However, detection small targets is difficult because they are more obscure and there are many target-like objects. Therefore, few studies have examined how to detect small targets consisting of fewer than $30{\times}10$ pixels. This paper presents a small target detection method using clutter rejection with stochastic hypothesis testing for FLIR imagery. The proposed algorithm consists of two stages; detection and clutter rejection. In the detection stage, the mean of the input FLIR image is first removed and then the image is segmented using Otsu's method. A closing operation is also applied during the detection stage in order to merge any single targets detected separately. Then, the residual of the clutters is eliminated using statistical hypothesis testing based on the t-test. Several FLIR images are used to prove the performance of the proposed algorithm. The experimental results show that the proposed algorithm accurately detects small targets (Jess than $30{\times}10$ pixels) with a low false alarm rate compared to the center-surround difference method using the receiver operating characteristics (ROC) curve.

• Fashion & Textile Research Journal
• /
• v.3 no.1
• /
• pp.61-69
• /
• 2001

The purpose of this study was to analyze the purchasing behavior related factors of Small Store User in Yhasi street of Dong-Sung Ro, Daegu. Frequency, $X^2$-test MANOVA, ANOVA and Duncan multiple range test were used to analyze the sample. The results of this study were as follows: 1) The largest sample were as follows: un married female, college students of twenties, 101-200 thousand won for salaries. 2) The factors of purchasing behavior were classified into 8 factors, enjoy shopping, store image, unique goods, culture space, salesperson, low price, information seeking, value via price orientation. 3) There were significant differences found between attitude on information source, number of seeking store, music in shop, music sound, size, display, price, street, in their factors of purchasing behavior (unique goods, value via price, low price, store image, enjoy shopping) 4) There were significant differences found between demographic characteristics (personal sales, location, transportation) in their factors of purchasing behavior (salesperson, cultural space, store image).

• Proceedings of the Korea Multimedia Society Conference
• /
• 2000.11a
• /
• pp.493-497
• /
• 2000

Fragile watermarking is a technique far autoentication/integrity of digital data. Unlike robust watermarking, il has to design to be vulnerable against some slight processing to verify the modification of digital data. Feature of fragile watermarking is to identify the modifications of data and to locate some places modification occurred at the same time, so it has to identify slight changes of small range if possible. In this paper, fragile watermarking is proposed that the changes of small range on image can be identified using the watermark sequence with period and the values of low bit planes in an image.

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2019.05a
• /
• pp.151-152
• /
• 2019

In this paper, we propose an improved learning method based on a small data set for animal image classification. First, CNN creates a training model for a small data set and uses the data set to expand the data set of the training set Second, a bottleneck of a small data set is extracted using a pre-trained network for a large data set such as VGG16 and stored in two NumPy files as a new training data set and a test data set, finally, learn the fully connected network as a new data set.

• Proceedings of the KIEE Conference
• /
• 2008.10b
• /
• pp.137-138
• /
• 2008

This paper proposes the modified image contrast enhancement technique for small-sized display of mobile handset. Sample images are user interface images, in which scaled up wVGA($800{\times}480$) from qVGA($320{\times}240$) that we can see easily in mobile handset. The display size of mobile handset is relatively small, so the goal of this paper is to simplify image contrast enhancement algorithm based on conventional HE (Histogram Equalization) algorithm and improve computational effectiveness to minimize power consumption in real hardware IC. In this paper, we adopt HE technique, which is classical and widely used for image contrast enhancement. At first, the input frame image is partitioned to temporal sub-frames and then analyzes gray level histogram of each sub-frame. In case that the analyzed histogram of some sub-frames deviates so much from reference level (it means that the sub-frame image components consist of too bright ones or dark ones), apply DHE(Dynamic Histogram Equalization) algorithm. In the other case, apply classical Histogram Linearization (or Global HE) algorithm. Also we compare the HE technique with gamma LUT (Look-Up Table) method, which is known as the simplest technique to enhance image contrast.

• The Journal of Korean Society for Radiation Therapy
• /
• v.25 no.2
• /
• pp.175-180
• /
• 2013

Purpose: Modern radiation therapy technique such as IGRT has become a routine clinical practice on LINAC for decrease patient's set-up error. CBCT can be used to adjust patient set-up error and treat patient more accurately. The Purpose of this study is to evaluate field size of CBCT for improving Image quality and suggest reference date of CBCT field size. Materials and Methods: Image date were acquired using KV CBCT and Catphan phantom (Half fan and full fan mode were scanned from 2 ~16 cm, at intervals of 2 cm). Field size were categorized by Small field size (2 cm, 4 cm), Medium field size (8 cm, 10 cm), Large field size (more than 14 cm) and evaluate. To estimated the CTDi using CTDi phantom and Ion chamber. Results: CT number linearity of Small and Large field size are greater than Medium field size. Spatial resolution are not significantly different without Small field size. But half fan mode is more different than full fan mode. In full fan, except Medium field size, all field size exceed recommendation for HU uniformity. But half pan has stability for all field except Small field size. CTDi makes radical sign function graph in Medium field size. Conclusion: The worst result was given by Small field size for Image quality and practically. Medium field size can be useful to prevent patient from radiation exposure and give better Image quality. So this study recommends that Medium field size (8~10 cm) is more suitable for CBCT.

• Journal of Broadcast Engineering
• /
• v.2 no.2
• /
• pp.156-169
• /
• 1997

This paper proposes an image segmentation technique using watershed algorithm followed by region merging method. A gradient image is obtained by applying multiscale gradient algorithm to the image simplified by morphological filters. Since the watershed algorithm produces the oversegmented image. it is necessary to merge small segmented regions as wel]' as region having similar characteristics. For region merging. we utilize the merging criteria based on both the mean value of the pixels of each region and the edge intensities between regions obtained by the contour following process. Experimental results show that the proposed method produces meaningful image segmentation results.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.21 no.4
• /
• pp.43-49
• /
• 2021

A small infrared image homing system is a tracking system that has an infrared image sensor that identifies a target through the day and night infrared image processing of the target on the ground and searches for and detects the target with respect to the main target. This paper describes the development of a board equipped with a high-speed CPU and FPGA (Field Programmable Gate Array) to identify target through real-time image processing by acquiring target information through infrared image. We propose a CPU-FPGA combining architecture for CPU and FPGA selection and video signal processing, and also describe a controller design using FPGA to control infrared sensor.