• 한국멀티미디어학회논문지
• /
• 제22권3호
• /
• pp.335-348
• /
• 2019

Cell segmentation and counting represent one of the most important tasks required in order to provide an exhaustive understanding of biological images. Conventional features suffer the lack of spatial consistency by causing the joining of the cells and, thus, complicating the cell counting task. We propose, in this work, a cascade of networks that take as inputs different versions of the original image. After constructing a Gaussian pyramid representation of the microscopy data, the inputs of different size and spatial resolution are given to a cascade of deep convolutional autoencoders whose task is to reconstruct the segmentation mask. The coarse masks obtained from the different networks are summed up in order to provide the final mask. The principal and main contribution of this work is to propose a novel method for the cell counting. Unlike the majority of the methods that use the obtained segmentation mask as the prior information for counting, we propose to utilize the hidden latent representations, often called the high-level features, as the inputs of a neural network based regressor. While the segmentation part of our method performs as good as the conventional deep learning methods, the proposed cell counting approach outperforms the state-of-the-art methods.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제14권11호
• /
• pp.4426-4442
• /
• 2020

Since scene segmentation is becoming a hot topic in the field of autonomous driving and medical image analysis, researchers are actively trying new methods to improve segmentation accuracy. At present, the main issues in image semantic segmentation are intra-class inconsistency and inter-class indistinction. From our analysis, the lack of global information as well as macroscopic discrimination on the object are the two main reasons. In this paper, we propose a Densely connected residual Attention network (DA-Res2Net) which consists of a dense residual network and channel attention guidance module to deal with these problems and improve the accuracy of image segmentation. Specifically, in order to make the extracted features equipped with stronger multi-scale characteristics, a densely connected residual network is proposed as a feature extractor. Furthermore, to improve the representativeness of each channel feature, we design a Channel-Attention-Guide module to make the model focusing on the high-level semantic features and low-level location features simultaneously. Experimental results show that the method achieves significant performance on various datasets. Compared to other state-of-the-art methods, the proposed method reaches the mean IOU accuracy of 83.2% on PASCAL VOC 2012 and 79.7% on Cityscapes dataset, respectively.

• 한국멀티미디어학회논문지
• /
• 제25권5호
• /
• pp.661-669
• /
• 2022

The purpose of this study is to compare and analyze Deep Learning (DL) and Digital Image Processing (DIP) techniques using the results of the glottis segmentation of the two methods followed by the quantification of the asymmetric degree of the vocal cord mucosa. The data consists of 40 normal and abnormal images. The DL model is based on Deeplab V3 architecture, and the Canny edge detector algorithm and morphological operations are used for the DIP technique. According to the segmentation results, the average accuracy of the DL model and the DIP was 97.5% and 94.7% respectively. The quantification results showed high correlation coefficients for both the DL experiment (r=0.8512, p<0.0001) and the DIP experiment (r=0.7784, p<0.0001). In the conclusion, the DL model showed relatively higher segmentation accuracy than the DIP. In this paper, we propose the clinical applicability of this technique applying the segmentation and asymmetric quantification algorithm to the glottal area in the laryngoscopic images.

• 대한의용생체공학회:의공학회지
• /
• 제42권5호
• /
• pp.201-210
• /
• 2021

In Korean medicine, tongue diagnosis is one of the important diagnostic methods for diagnosing abnormalities in the body. Representative features that are used in the tongue diagnosis include color, shape, texture, cracks, and tooth marks. When diagnosing a patient through these features, the diagnosis criteria may be different for each oriental medical doctor, and even the same person may have different diagnosis results depending on time and work environment. In order to overcome this problem, recent studies to automate and standardize tongue diagnosis using machine learning are continuing and the basic process of such a machine learning-based tongue diagnosis system is tongue segmentation. In this paper, image data is augmented based on the main tongue features, and backbones of various famous deep learning architecture models are used for automatic tongue segmentation. The experimental results show that the proposed augmentation technique improves the accuracy of tongue segmentation, and that automatic tongue segmentation can be performed with a high accuracy of 99.12%.

• ETRI Journal
• /
• 제45권5호
• /
• pp.822-835
• /
• 2023

Recent semantic segmentation frameworks usually combine low-level and high-level context information to achieve improved performance. In addition, postlevel context information is also considered. In this study, we present a Context ReFinement Network (CRFNet) and its training method to improve the semantic predictions of segmentation models of the encoder-decoder structure. Our study is based on postprocessing, which directly considers the relationship between spatially neighboring pixels of a label map, such as Markov and conditional random fields. CRFNet comprises two modules: a refiner and a combiner that, respectively, refine the context information from the output features of the conventional semantic segmentation network model and combine the refined features with the intermediate features from the decoding process of the segmentation model to produce the final output. To train CRFNet to refine the semantic predictions more accurately, we proposed a sequential training scheme. Using various backbone networks (ENet, ERFNet, and HyperSeg), we extensively evaluated our model on three large-scale, real-world datasets to demonstrate the effectiveness of our approach.

• 대한원격탐사학회지
• /
• 제23권6호
• /
• pp.521-528
• /
• 2007

본 논문에서는 대상지역에 대한 영상을 다양한 가중치의 조합의 경우를 고려하여 객체 단위로 분할하게 되며 분할된 객체에 대하여 상호관계를 분석하여 수치적으로 표현하였다. 또한 최종적인 객체 기반영상분류에서 높은 정확도를 확보할 수 있는 가중치의 조합을 산정하였다. 연구에 사용된 영상은 Landsat-7/ETM 영상으로 대상 지역의 면적은 $11{\times}14$ Km이며 밴드 2, 3, 4의 조합을 사용하였다. 객체 간 계산은 Moran's I와 객체 내부 분산(Intrasegment Variance)을 이용하였다. 대상지역에 대하여 총 75개의 가중치 조합을 사용하여 75개의 객체 분할 영상을 생성하였다. 객체 분할 영상 중에 최종적인 영상 분류 시 높은 정확도가 예상되는 가중치 조합, 중간 정도 정확도가 예상되는 가중치 조합 그리고 낮은 정도 정확도가 예상되는 가중치 조합을 7개 선택하여 최종적인 객체기반 영상분류를 시행하고 그 정확도를 비교하였다. 정확도의 비교 결과, 가장 높은 정확도가 예상되는 가중치 조합의 객체 분할 영상의 경우 객체 기반 영상 분류 시 85% 이상의 정확도를 나타내었으며 반대로 낮은 경우는 분류 시 50% 정도의 분류 정확도를 나타내었다.

• 대한원격탐사학회지
• /
• 제38권6_1호
• /
• pp.1091-1100
• /
• 2022

딥러닝 기술의 진보와 함께 다양한 국내외 고해상도 원격탐사 영상의 활용이 가능함에 따라 딥러닝 기술과 원격탐사 빅데이터를 활용하여 도심 지역 건물 검출과 변화탐지에 활용하고자 하는 관심이 크게 증가하고 있다. 본 논문에서는 고해상도 원격탐사 영상의 의미론적 건물 분할을 위해서 건물 분할에 우수한 성능을 보이는 DeepResUNet 모델을 기본 구조로 하고 잔차 학습 단위를 개선하고 Convolutional Block Attention Module(CBAM)을 결합한 새로운 건물 분할 모델인 CBAM-DRUNet을 제안한다. 제안한 건물 분할 모델은 WHU 데이터셋과 INRIA 데이터셋을 이용한 성능 평가에서 UNet을 비롯하여 ResUNet, DeepResUNet 대비 F1 score, 정확도, 재현율 측면에서 모두 우수한 성능을 보였다.

• Korean Journal of Radiology
• /
• 제22권6호
• /
• pp.901-911
• /
• 2021

Objective: To evaluate the technical applicability of a semiautomatic three-dimensional (3D) hybrid CT segmentation method for the quantification of right ventricular mass in patients with cardiovascular disease. Materials and Methods: Cardiac CT (270 cardiac phases) was used to quantify right ventricular mass using a semiautomatic 3D hybrid segmentation method in 195 patients with cardiovascular disease. Data from 270 cardiac phases were divided into subgroups based on the extent of the segmentation error (no error; ≤ 10% error; > 10% error [technical failure]), defined as discontinuous areas in the right ventricular myocardium. The reproducibility of the right ventricular mass quantification was assessed. In patients with no error or < 10% error, the right ventricular mass was compared and correlated between paired end-systolic and end-diastolic data. The error rate and right ventricular mass were compared based on right ventricular hypertrophy groups. Results: The quantification of right ventricular mass was technically applicable in 96.3% (260/270) of CT data, with no error in 54.4% (147/270) and ≤ 10% error in 41.9% (113/270) of cases. Technical failure was observed in 3.7% (10/270) of cases. The reproducibility of the quantification was high (intraclass correlation coefficient = 0.999, p < 0.001). The indexed mass was significantly greater at end-systole than at end-diastole (45.9 ± 22.1 g/m² vs. 39.7 ± 20.2 g/m², p < 0.001), and paired values were highly correlated (r = 0.96, p < 0.001). Fewer errors were observed in severe right ventricular hypertrophy and at the end-systolic phase. The indexed right ventricular mass was significantly higher in severe right ventricular hypertrophy (p < 0.02), except in the comparison of the end-diastolic data between no hypertrophy and mild hypertrophy groups (p > 0.1). Conclusion: CT quantification of right ventricular mass using a semiautomatic 3D hybrid segmentation is technically applicable with high reproducibility in most patients with cardiovascular disease.

• 한국HCI학회:학술대회논문집
• /
• 한국HCI학회 2009년도 학술대회
• /
• pp.193-196
• /
• 2009

단일 영상에서 피부색 영역을 추출하기 위해서 기존의 많은 방법들이 하나의 고정된 피부색 모델을 사용한다. 그러나 영상에 특성에 따라 영상에 포함된 피부색의 분포가 다양하기 때문에 이러한 방법을 이용하여 피부색을 검출할 경우 낮은 검출율이나 높은 긍정 오류율이 발생할 수 있다. 따라서 영상의 특징에 따라 적응적으로 피부색 영역을 추출할 수 있는 방법이 필요하다. 이에 본 논문에서는 영상의 특징에 따라 2단계의 과정을 거쳐 피부색 모델을 수정하는 방법으로, 다양한 조명과 환경 조건에서 높은 검출율과 낮은 긍정 오류율을 동시에 가지는 알고리즘을 제안한다.

• 대한의용생체공학회:의공학회지
• /
• 제28권3호
• /
• pp.332-337
• /
• 2007

In the early twenty-first century, minimally invasive surgery is the mainstay of various kinds of surgical fields. Surgeons gave percutaneously surgical treatment of the screw directly using a fluoroscopic view in the past. The latest date, they began to operate the fractured carpal bone surgery using Computerized Tomography (CT). Carpal bones composed of wrist joint consist of eight small bones which have hexahedron and sponge shape. Because of these shape, it is difficult to grasp the shape of carpal bones using only CT image data. Although several image segmentation studies have been conducted with carpal bone CT image data, more studies about carpal bone using CT data are still required. Especially, to apply the software implemented from the studies to clinical fIeld, the outcomes should be user friendly and very accurate. To satisfy those conditions, we propose modified multi-seed region growing segmentation method which uses simple threshold and the canny edge detector for finding edge information more accurately. This method is able to use very easily and gives us high accuracy and high speed for extracting the edge information of carpal bones. Especially, using multi-seed points, multi-bone objects of the carpal bone are extracted simultaneously.