• 대한의용생체공학회:의공학회지
• /
• 제43권5호
• /
• pp.341-352
• /
• 2022

Medical image segmentation is the most important task in radiation therapy. Especially, when segmenting medical images, the liver is one of the most difficult organs to segment because it has various shapes and is close to other organs. Therefore, automatic segmentation of the liver in computed tomography (CT) images is a difficult task. Since tumors also have low contrast in surrounding tissues, and the shape, location, size, and number of tumors vary from patient to patient, accurate tumor segmentation takes a long time. In this study, we propose a method algorithm for automatically segmenting the liver and tumor for this purpose. As an advantage of setting the boundaries of the tumor, the liver and tumor were automatically segmented from the CT image using the 2D CoordConv DeepLab V3+ model using the CoordConv layer. For tumors, only cropped liver images were used to improve accuracy. Additionally, to increase the segmentation accuracy, augmentation, preprocess, loss function, and hyperparameter were used to find optimal values. We compared the CoordConv DeepLab v3+ model using the CoordConv layer and the DeepLab V3+ model without the CoordConv layer to determine whether they affected the segmentation accuracy. The data sets used included 131 hepatic tumor segmentation (LiTS) challenge data sets (100 train sets, 16 validation sets, and 15 test sets). Additional learned data were tested using 15 clinical data from Seoul St. Mary's Hospital. The evaluation was compared with the study results learned with a two-dimensional deep learning-based model. Dice values without the CoordConv layer achieved 0.965 ± 0.01 for liver segmentation and 0.925 ± 0.04 for tumor segmentation using the LiTS data set. Results from the clinical data set achieved 0.927 ± 0.02 for liver division and 0.903 ± 0.05 for tumor division. The dice values using the CoordConv layer achieved 0.989 ± 0.02 for liver segmentation and 0.937 ± 0.07 for tumor segmentation using the LiTS data set. Results from the clinical data set achieved 0.944 ± 0.02 for liver division and 0.916 ± 0.18 for tumor division. The use of CoordConv layers improves the segmentation accuracy. The highest of the most recently published values were 0.960 and 0.749 for liver and tumor division, respectively. However, better performance was achieved with 0.989 and 0.937 results for liver and tumor, which would have been used with the algorithm proposed in this study. The algorithm proposed in this study can play a useful role in treatment planning by improving contouring accuracy and reducing time when segmentation evaluation of liver and tumor is performed. And accurate identification of liver anatomy in medical imaging applications, such as surgical planning, as well as radiotherapy, which can leverage the findings of this study, can help clinical evaluation of the risks and benefits of liver intervention.

• 한국측량학회지
• /
• 제20권1호
• /
• pp.21-31
• /
• 2002

이 연구는 원격탐사 영상의 분류정확도를 향상시키기 위한 방법으로써 신경망 이론과 퍼지집합이론을 각각 적용하였다. 원격탐사 영상은 토지피복도, 식생도, 지질도 등 주제도를 만드는데 많이 이용되고 있다. 원격탐사 영상의 감독분류에 대한 정확도는 트레이닝 지역의 선정, 분류항목의 할당 문제로 인해 많은 차이를 보인다. 일반적인 영상 분류법은 영상 내의 모든 영상소가 균질하다고 가정한다. 그러나, 이러한 가정은 영상내의 수많은 혼합 영상소를 분류해내는 데에는 적합하지 않다. 이러한 문제를 극복하기 위해 퍼지 집합이론을 적용하였으며, 퍼지 집합이론의 멤버쉽을 이용하였다. 퍼지 집합이론은 하나의 영상소를 멤버쉽의 정도에 따라 여러 가지 항목으로 분류할 수 있는 장점이 있다. 그러나, 퍼지분류법과 통계학적인 분류법은 화소값의 분포가 비정규적일 때 좋지 않은 분류 결과를 나타내며 처리 시간이 늦고 많은 컴퓨팅 비용이 드는 단점이 있다. 그 대안적인 방법으로서 신경망분류법을 들 수 있는데, 신경망 분류법은 비모수적 분류법으로서 일반적인 분류기법보다 좀 더 좋은 결과를 나타내고 있고, 한번 트레이닝 되면 빠르게 데이터를 분류할 수 있다.

• 한국정보처리학회논문지
• /
• 제7권2호
• /
• pp.365-372
• /
• 2000

프랙탈 영상 압축(Fractral Image Compression:FIC)의 진화 계산(Evolution Computation)을 이용한 영상 분할(Image Partition)을 소개한다. 프랙탈 영상 압축에서 지역(Ranges)의 영상 분할은 꼭 필요하다[1]. 프랙탈 영상 압축은 쉽고 빠르게 복원된다는 장점을 갖는 데 비해 반복적인 프랙탈 변환의 적용으로 많은 계산량을 필요로 한다는 단점을 가지고 있다. 위와 같은 문제점을 해결하기 위한 방법으로 영상 분할을 하는데 있어 진화 계산을 적용하는 것에 대해 제안한다. 치역 영상(Ranges Image)은 작은 사각(Square) 영상 블록들의 결합된 집합으로 구성할 수 있다. 모집단을 구성하는 하나의 $N_p$는 분할되어진 하나의 코드들이다. 진화 계산에서 각각의 구성은 두 개의 이웃하는 치역은 제외하고 그들의 부모(Parent)로부터 분할을 상속받은 자식 $\sigma$를 생성한다. 자손들의 최적의 영상은 콜라주 정리(Collage Theorem)에 기초를 둔 다음 세대 모집단을 위해 선택되어지고 처리된다. 최적의 영상은 영상 데이터에 포함된 중복성을 포함함으로서 적은 저장 공간을 차지하고 속도 문제에 있어서 효율적이고 영상의 화질에 있어서 다른 부호화를 사용한 기법보다 우수한 성능을 갖는다. 멀티미디어 영상 처리(Multimedia Image Processing)의 진화 계산을 이용한 프렉탈 영상 압축은 영상의 복원과 영상의 질, 고 압축률을 요하는 동영상의 적용등의 많은 분야에 적용된다.

• 복식문화연구
• /
• 제16권5호
• /
• pp.863-877
• /
• 2008

The purpose of this study is to investigate the effect of dot pattern size(0.8, 1.8, 2.5, 5, 8), color combination (BG/R, Y/B), value tone(lt/dk, p/g), area-ratio on image information. Sets of stimulus and response scales(7 point semantic) were used as experimental materials. The stimuli were 20 color pictures manipulated with the combination of dot pattern size, color combination, value tone and area-ratio using computer simulation. The subjects were 240 female undergraduates living in Gyeongsangnam-do. Image factor of the stimulus was composed of 4 different components, visibility, chastity.feminity, cuteness and attractiveness. In the visibility, color combination, value tone, area-ratio, dot pattern size showed independent effect. In the chastity feminity, color combination, value tone, showed independent effect. In the cuteness, value tone, area-ratio, dot pattern size showed independent effect. Significant interaction effects of color and area-ratio combination on visibility and cuteness were found. Interaction efforts of color and value tone combination, value tone and area-ratio was significant on cuteness. For visibility image, BG/R combination of color and yellow background/blue dots were effective. For cuteness image, pale/grayish tone and background/dots area-ratio were effective.

• 복식
• /
• 제57권5호
• /
• pp.72-89
• /
• 2007

The purpose of this study is to investigate the effect of eyeshadow color(brown, purple), lipstick color(red, red purple, and yellow red), and lipstick tone(vivid, light, dull, and dark), clothing color(same, different), clothing tone(vivid, light, dull, and dark) on image formation. Sets of stimulus and response scales(7 point semantic) were used as experimental materials. The stimuli were 128 color pictures manipulated with the combination of eyeshadow color, lipstick color, lipstick tone, clothing color, and clothing tone using computer simulation. The subjects were 768 female undergraduates living in Gyeongnam-do. Image factor of the stimulus was composed of 4 different components, attractiveness, visibility, stability, and softness). In the 4 image components, lipstick tone, clothing color and clothing tone showed independent effect. Eyeshadow color influenced independently on the attractiveness, visibility and stability. According to the variation of clothing color and tone, makeup color, it was investigated that the images for a clothing wearer were expressed diversely, were shown differently in image dimensions, and could be produced to different images.

• 한국생활과학회지
• /
• 제15권4호
• /
• pp.611-622
• /
• 2006

The purpose of this study is to investigate the effect of eyeshadow color (brown, purple, and blue), lipstick color (red, orange, and purple), and lipstick tone(vivid, light, dull, and dark) on the makeup image. The experimental materials used for this study were sets of stimulus and response scales (7 point semantic). The stimuli were 36 color pictures manipulated with the combination of eyeshadow color, lipstick color, and lipstick tone using computer simulation. The subjects were 216 female undergraduates living in Jinju city. The data was analyzed by using SPSS program. Analyzing methods were ANOVA and Duncan test. The result of this study are as follows. Image factor of the stimulus was composed of 4 different components (attractiveness and gracefulness, visibility, cuteness, and softness), Among them, the attractiveness and gracefulness and the visibility were important. Each dimensional image was affected by color coordination of eyeshadow color, lipstick color and lipstick tone. Therefore, the face image through matching eyeshadow and lipstick could be varied by the eyeshadow color, lipstick color and tones.

• 대한원격탐사학회:학술대회논문집
• /
• 대한원격탐사학회 2006년도 Proceedings of ISRS 2006 PORSEC Volume I
• /
• pp.260-262
• /
• 2006

One of future remote sensing techniques for transportation application is vehicle detection from the space, which could be the basis of measuring traffic volume and recognizing traffic condition in the future. This paper introduces an approach to vehicle detection using image object segmentation approach. The object-oriented image processing is particularly beneficial to high-resolution image classification of urban area, which suffers from noisy components in general. The project site was Dae-Jeon metropolitan area and a set of true color aerial images at 10cm resolution was used for the test. Authors investigated a variety of parameters such as scale, color, and shape and produced a customized solution for vehicle detection, which is based on a knowledge-based hierarchical model in the environment of eCognition. The highest tumbling block of the vehicle detection in the given data sets was to discriminate vehicles in dark color from new black asphalt pavement. Except for the cases, the overall accuracy was over 90%.

• JSTS:Journal of Semiconductor Technology and Science
• /
• 제3권2호
• /
• pp.83-88
• /
• 2003

The reset operation of a CCD image sensor was improved using charge trapping of a MOS structure to realize a loe voltage driving. A DC bias generating circuit was added to the reset structure which sets reference voltage and holds the signal charge to be detected. The generated DC bias is added to the reset pulse to give an optimized voltage margin to the reset operation, and is controlled by adjustment of the threshold voltage of a MOS transistor in the circuit. By the pulse-type stress voltage applied to the gate, the electrons and holes were injected to the gate dielectrics, and the threshold voltage could be adjusted ranging from 0.2V to 5.5V, which is suitable for controlling the incomplete reset operation due to the process variation. The charges trapped in the silicon nitride lead to the positive and negative shift of the threshold voltage, and this phenomenon is explained by Poole-Frenkel conduction and Fowler-Nordheim conduction. A CCD image sensor with $492(H){\;}{\times}{\;}510(V)$ pixels adopting this structure showed complete reset operation with the driving voltage of 3.0V. The resolution chart taken with the image sensor shows no image flow to the illumination of 30 lux, even in the driving voltage of 3.0V.

• International Journal of Computer Science & Network Security
• /
• 제23권11호
• /
• pp.93-98
• /
• 2023

Instagram is one of the fastest-growing online photo social web services where users share their life images and videos with other users. Image tagging is an essential step for developing Automatic Image Annotation (AIA) methods that are based on the learning by example paradigm. Hashtags can be used on just about any social media platform, but they're most popular on Twitter and Instagram. Using hashtags is essentially a way to group together conversations or content around a certain topic, making it easy for people to find content that interests them. Practically on average, 20% of the Instagram hashtags are related to the actual visual content of the image they accompany, i.e., they are descriptive hashtags, while there are many irrelevant hashtags, i.e., stophashtags, that are used across totally different images just for gathering clicks and for search ability enhancement. Hence in this work, Sorting instagram hashtags all the way through mass tagging using HITS (Hyperlink-Induced Topic Search) algorithm is presented. The hashtags can sorted to several groups according to Jensen-Shannon divergence between any two hashtags. This approach provides an effective and consistent way for finding pairs of Instagram images and hashtags, which lead to representative and noise-free training sets for content-based image retrieval. The HITS algorithm is first used to rank the annotators in terms of their effectiveness in the crowd tagging task and then to identify the right hashtags per image.

• 한국멀티미디어학회논문지
• /
• 제21권2호
• /
• pp.98-107
• /
• 2018

In this paper, we propose a method to classify the skeletal maturity with a small amount of hand wrist X-ray image using deep learning-based meta-learning. General deep-learning techniques require large amounts of data, but in many cases, these data sets are not available for practical application. Lack of learning data is usually solved through transfer learning using pre-trained models with large data sets. However, transfer learning performance may be degraded due to over fitting for unknown new task with small data, which results in poor generalization capability. In addition, medical images require high cost resources such as a professional manpower and mcuh time to obtain labeled data. Therefore, in this paper, we use meta-learning that can classify using only a small amount of new data by pre-trained models trained with various learning tasks. First, we train the meta-model by using a separate data set composed of various learning tasks. The network learns to classify the bone maturity using the bone maturity data composed of the radiographs of the wrist. Then, we compare the results of the classification using the conventional learning algorithm with the results of the meta learning by the same number of learning data sets.