• The Journal of Korean Institute of Next Generation Computing
• /
• v.15 no.5
• /
• pp.64-74
• /
• 2019

Although a non-rigid registration has high demands in clinical practice, it has a high computational complexity and it is very difficult for ensuring the accuracy and robustness of registration. This study proposes a method of applying a non-rigid registration to 3D magnetic resonance images of brain in an unsupervised learning environment by using a deep-learning network. A feature vector between two images is produced through the network by receiving both images from two different patients as inputs and it transforms the target image to match the source image by creating a displacement vector field. The network is designed based on a U-Net shape so that feature vectors that consider all global and local differences between two images can be constructed when performing the registration. As a regularization term is added to a loss function, a transformation result similar to that of a real brain movement can be obtained after the application of trilinear interpolation. This method enables a non-rigid registration with a single-pass deformation by only receiving two arbitrary images as inputs through an unsupervised learning. Therefore, it can perform faster than other non-learning-based registration methods that require iterative optimization processes. Our experiment was performed with 3D magnetic resonance images of 50 human brains, and the measurement result of the dice similarity coefficient confirmed an approximately 16% similarity improvement by using our method after the registration. It also showed a similar performance compared with the non-learning-based method, with about 10,000 times speed increase. The proposed method can be used for non-rigid registration of various kinds of medical image data.

• IEIE Transactions on Smart Processing and Computing
• /
• v.6 no.4
• /
• pp.262-268
• /
• 2017

Intelligent video surveillance systems have been developed to monitor global areas and find specific target objects using a large-scale database. However, person re-identification presents some challenges, such as pose change and occlusions. To solve the problems, this paper presents an improved person re-identification method using sparse representation and saliency-based dictionary construction. The proposed method consists of three parts: i) feature description based on salient colors and textures for dictionary elements, ii) orthogonal atom selection using cosine similarity to deal with pose and viewpoint change, and iii) measurement of reconstruction error to rank the gallery corresponding a probe object. The proposed method provides good performance, since robust descriptors used as a dictionary atom are generated by weighting some salient features, and dictionary atoms are selected by reducing excessive redundancy causing low accuracy. Therefore, the proposed method can be applied in a large scale-database surveillance system to search for a specific object.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.37 no.3
• /
• pp.79-87
• /
• 2000

In this paper, we propose a model based moving object tracking algorithm In dynamic scenes To adapt shape change of the moving object, the Hausdorff distance is applied as the measurement of similarity between model and image To reduce processing time, 2D logarithmic search method is applied for locate the position of moving object Experiments on a running vehicle and motorcycle, the result showed that the mean square error of real position and tracking result is 1150 and 1845; matching times are reduced average 1125times and 523 times than existing algorithm for vehicle image and motorcycle image, respectively It showed that the proposed algorithm could track the moving object accurately.

• Journal of Information Processing Systems
• /
• v.16 no.3
• /
• pp.663-676
• /
• 2020

Plant leaf classification is a significant application of image processing techniques in modern agriculture. In this paper, a multi-granular angle description method is proposed for plant leaf classification and retrieval. The proposed method can describe leaf information from coarse to fine using multi-granular angle features. In the proposed method, each leaf contour is partitioned first with equal arc length under different granularities. And then three kinds of angle features are derived under each granular partition of leaf contour: angle value, angle histogram, and angular ternary pattern. These multi-granular angle features can capture both local and globe information of the leaf contour, and make a comprehensive description. In leaf matching stage, the simple city block metric is used to compute the dissimilarity of each pair of leaf under different granularities. And the matching scores at different granularities are fused based on quotient space theory to obtain the final leaf similarity measurement. Plant leaf classification and retrieval experiments are conducted on two challenging leaf image databases: Swedish leaf database and Flavia leaf database. The experimental results and the comparison with state-of-the-art methods indicate that proposed method has promising classification and retrieval performance.

• Journal of the Korea Society of Computer and Information
• /
• v.24 no.3
• /
• pp.41-47
• /
• 2019

In this paper, we propose a application of conditional generative adversarial network (cGAN) for generation of contrast enhanced computed tomography (CT) image. Two types of CT data which were the enhanced and non-enhanced were used and applied by the histogram equalization for adjusting image intensities. In order to validate the generation of contrast enhanced CT data, the structural similarity index measurement (SSIM) was performed. Prepared generated contrast CT data were analyzed the statistical analysis using paired sample t-test. In order to apply the optimized algorithm for the lymph node cancer, they were calculated by short to long axis ratio (S/L) method. In the case of the model trained with CT data and their histogram equalized SSIM were $0.905{\pm}0.048$ and $0.908{\pm}0.047$. The tumor S/L of generated contrast enhanced CT data were validated similar to the ground truth when they were compared to scanned contrast enhanced CT data. It is expected that advantages of Generated contrast enhanced CT data based on deep learning are a cost-effective and less radiation exposure as well as further anatomical information with non-enhanced CT data.

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.47 no.1
• /
• pp.73-79
• /
• 2010

Emotion intelligence computing is able to processing of human emotion through it's studying and adaptation. Also, Be able more efficient to interaction of human and computer. As sight and hearing, music & image is constitute of short time and continue for long. Cause to success marketing, understand-translate of humanity emotion. In this paper, Be design of check system that matched music and image by user emotion keyword(irritability, gloom, calmness, joy). Suggested system is definition by 4 stage situations. Then, Using music & image and emotion ontology to retrieval normalized music & image. Also, A sampling of image peculiarity information and similarity measurement is able to get wanted result. At the same time, Matched on one space through pared correspondence analysis and factor analysis for classify image emotion recognition information. Experimentation findings, Suggest system was show 82.4% matching rate about 4 stage emotion condition.

• Journal of the Korean Society for Aeronautical & Space Sciences
• /
• v.37 no.12
• /
• pp.1165-1172
• /
• 2009

The laminar boundary layer along a streamwise corner formed by two flat plates intersecting at right angle is measured by using Particle Image Velocimetry(PIV) technique. The free stream velocity ranges from 2.96m/s to 3.0m/s. The angle of incidence of the corner is set to 1.2 degree providing slightly favourable pressure gradient to ensure a laminar flow in the corner region. A round shape leading edge is used and the length of the model is about 1000mm. In the bisector plane, the measurement data show separation type velocity profiles having an inflection point which is a typical characteristic of laminar corner boundary layers. As the distance away from the bisector plane increases, velocity profiles are found to change into the Blasius profile. The change completes around half length of the boundary layer thickness in the bisector plane away from the bisector plane along the plate. In the bisector plane, the growth characteristic of the boundary layer thickness and the approximate similarity of velocity profiles are confirmed from the measurement data.

• Proceedings of the Korean Society for Agricultural Machinery Conference
• /
• 2017.04a
• /
• pp.157-157
• /
• 2017

To minimize the damage by wild birds and acquire the benefits such as protection against weeds and maintenance of water content in soil, the mulching black color vinyl after seeding should be carried out. Non-contact and non-destructive methods that can continuously determine the locations are necessary. In this study, a crop position detection method was studied that uses infrared thermal image sensor to determine the cotyledon position under vinyl mulch. The moving system for acquiring image arrays has been developed for continuously detecting crop locations under plastic mulching on the field. A sliding mechanical device was developed to move the sensor, which were arranged in the form of a linear array, perpendicular to the array using a micro-controller integrated with a stepping motor. The experiments were conducted while moving 4.00 cm/s speed of the IR sensor by the rotational speed of the stepping motor based on a digital pulse width modulation signal from the micro-controller. The acquired images were calibrated with the spatial image correlation. The collected data were processed using moving averaging on interpolation to determine the frame where the variance was the smallest in resolution units of 1.02 cm. Non-linear integral interpolation was one of method for analyzing the frequency using the normalization image and then arbitrarily increasing the limited data value of $16{\times}4pixels$ in one frame. It was a method to relatively reduce the size of overlapping pixels by arbitrarily increasing the limited data value. The splitted frames into 0.1 units instead of 1 pixel can propose more than 10 times more accurate and original method than the existing correction method. The non-integral calibration method was conducted by applying the subdivision method to the pixels to find the optimal correction resolution based on the first reversed frequency. In order to find a correct resolution, the expected location of the first crop was indicated on near pixel 4 in the inversion frequency. For the most optimized resolution, the pixel was divided by 0.4 pixel instead of one pixel to find out where the lowest frequency exists.

• The Korean Journal of Nuclear Medicine Technology
• /
• v.16 no.2
• /
• pp.99-105
• /
• 2012

Purpose : This study is designed to compare two parameters reflecting $^{18}F$-FDG uptake, SUV and radioactivity, for diagnosis of thyroid cancer in dual time $^{18}F$-FDG PET/CT imaging and to find which parameter is more useful to decide whether the tumor is malignant or not. Materials and Methods : We performed retrospective study for 40 patients. All patients are diagnosed as primary thyroid cancer and examined $^{18}F$-FDG PET/CT. First, we got the dispersion of scattering beam of neck and lung apex to set a background and compared each dispersion, mean value, standard deviation of maxSUV and radioactivity. Also, mean maxSUV, ${\Delta}maxSUV$, ${\Delta}maxBq$/ml(%) and radioactivity between groups according to lesion's size based on biopsy are compared with independent-sample t-test. Results : the values that were from maxSUV and radioactivity measurement technique were compensated and calculated to practical values for mean comparison and patients were divided to two groups based on tumor size, Group1 ($size{\leq}1$ cm, n=21), Group2 (size>1 cm, n=19) for accurate comparison. In Group1, maxSUV (semi-quantitative analysis) was increased from $5.64{\pm}5.85$ (1.89~17.84) at first image to $5.90{\pm}5.01$ (1.95~18.22) at second image and radioactivity (Bq/ml) (quantitative analysis) showed similar increase from $5.93{\pm}6.38$ (2.50~16.75) at first image to $6.01{\pm}5.25$ (2.66~16.58) at second image. In Group2, TFmaxSUV was $10.54{\pm}14.36$ (2.54~33.89) in true first image, TSmaxSUV was $9.85{\pm}12.88$ (2.62~26.20) in true second image separately. The maxSUV showed a significant difference in the mean comparison between the two groups (p=0.035) But, mean radioactivity (Bq/ml) was $5.93{\pm}6.38$ (4.81~40.99) in true first image, $6.01{\pm}5.25$ (4.51~36.93) in true second image and didn't show a significant difference statistically (p=0.126) Conclusion : In diagnosis of thyroid tumor, SUV and radioactivity depending on $^{18}F$-FDG uptake showed high similarity with coefficient of determination (R2=0.939) and malignant evaluation results using dual time also showed similar aspect. Radioactivity for evaluation of malignant tumor didn't show better specificity or sensitivity than maxSUV.

• Proceedings of the Korean Information Science Society Conference
• /
• 2003.04c
• /
• pp.253-255
• /
• 2003

기존의 개념 기반 이미지 검색에서는 이미지의 의미적 내용 인식을 위해 일반적으로 어휘적 정보나 텍스트 정보를 이용했다. 이러한 텍스트 정보 기반 이미지 검색은 전통적인 검색 방법인 키워드 검색 기술을 그대로 사용하여 쉽게 구현할 수 있으나 텍스트의 개념적 매칭이 아닌 스트링 매칭이므로 주석처리된 단어와 정확한 매칭이 없다면 찾을 수가 없었다. 이에 본 논문에서는 ontology의 일종인 WordNet을 이용하여 깊이 정보량 링크 타입, 밀도 등을 고려한 개념간 유사성 측정으로 패턴 매칭의 문제를 해결하고자 했다. 또한 키워드로 주석처리 되어 있는 Microsofts Design Gallery Live의 이미지를 이용하여 개념간 유사성 측정법을 실질적으로 개념 기반 이미지 검색에 적용해 보았다.