• Journal of the Korean Society for Industrial and Applied Mathematics
• /
• 제27권2호
• /
• pp.135-145
• /
• 2023

Recently, as deep learning technology has developed, various deep learning technologies have been introduced in handwritten recognition, greatly contributing to performance improvement. The recognition accuracy of handwritten Hangeul recognition has also improved significantly, but prior research has focused on recognizing 520 Hangul characters or 2,350 Hangul characters using SERI95 data or PE92 data. In the past, most of the expressions were possible with 2,350 Hangul characters, but as globalization progresses and information and communication technology develops, there are many cases where various foreign words need to be expressed in Hangul. In this paper, we propose a model that recognizes and combines the consonants, medial vowels, and final consonants of a Korean syllable using a multi-label classification model, and achieves a high recognition accuracy of 98.38% as a result of learning with the public data of Korean handwritten characters, PE92. In addition, this model learned only 2,350 Hangul characters, but can recognize the characters which is not included in the 2,350 Hangul characters

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제13권11호
• /
• pp.5546-5559
• /
• 2019

Facial action units (AUs) have recently drawn increased attention because they can be used to recognize facial expressions. A variety of methods have been designed for frontal-view AU detection, but few have been able to handle multi-view face images. In this paper we propose a method for multi-view facial AU detection using a fused multilayer, multi-task, and multi-label deep learning network. The network can complete two tasks: AU detection and facial view detection. AU detection is a multi-label problem and facial view detection is a single-label problem. A residual network and multilayer fusion are applied to obtain more representative features. Our method is effective and performs well. The F1 score on FERA 2017 is 13.1% higher than the baseline. The facial view recognition accuracy is 0.991. This shows that our multi-task, multi-label model could achieve good performance on the two tasks.

• 한국전자거래학회지
• /
• 제26권1호
• /
• pp.29-41
• /
• 2021

최근 패션업계에서는 급속도로 발전하는 딥러닝 방법론을 활용하려는 시도가 늘고 있다. 이에 따라 다양한 패션 관련 문제들을 다루는 연구들이 제안되었고, 우수한 성능을 달성하였다. 하지만 패션 스타일 분류 문제의 경우, 기존 연구들은 한 옷차림이 여러 스타일을 동시에 포함할 수 있다는 패션 스타일의 특성을 반영하지 못하였다. 따라서 본 연구에서는 동시에 존재하는 레이블 간의 종속성을 모델링하고, 이를 반영하여 패션 스타일의 다중 분류 문제를 해결하고자 한다. 패션 스타일 사이의 종속성을 포착하고 탐색하기 위해 GCN(graph convolution network) 기반의 다중 레이블 인식 모델을 적용하였다. 또한 전이학습을 통해 모델의 학습 속도 및 성능을 향상시켰다. 제안하는 모델은 웹 크롤링을 통해 수집한 SNS 이미지 데이터를 이용하여 검증하였으며, 비교 모델 대비 우수한 성능을 기록하였다.

• 정보처리학회논문지:소프트웨어 및 데이터공학
• /
• 제10권12호
• /
• pp.555-560
• /
• 2021

최근 안면 피부 미용에 대한 사람들의 관심이 높아짐에 따라 딥 러닝을 활용한 안면 피부 미용을 위한 피부 질환 인식 연구가 진행되고 있다. 이러한 연구들은 여드름을 비롯한 다양한 피부 질환을 인식한다. 기존의 연구들은 단일 피부 질환만을 인식하지만, 안면에 발생하는 피부 질환은 더 다양하고 복합적으로 발생할 수 있다. 따라서 본 논문에서는 Inception-ResNet V2 모델을 활용하여 다중 레이블 분류 방법으로 여드름, 블랙헤드, 주근깨, 검버섯, 일반 피부, 화이트헤드에 관한 복합적인 피부 질환을 인식한다. 사용한 평가 지표 중 정확도는 98.8%, 해밍 손실은 0.003을 달성하였고, 단일 클래스별 정밀도, 재현율, F1-점수는 모두 96.6% 이상을 달성하였다.

• International Journal of Contents
• /
• 제6권1호
• /
• pp.6-11
• /
• 2010

Recently, there has been research to use portable digital camera to recognize objects in natural scene images, including labels or marks on a cylindrical surface. In many cases, text or logo in a label can be distorted by a structural movement of the object on which the label resides. Since the distortion in the label can degrade the performance of object recognition, the label should be rectified or restored from deformations. In this paper, a new method for label detection and restoration in digital images is presented. In the detection phase, the Hough transform is employed to detect two vertical boundaries of the label, and a horizontal edge profile is analyzed to detect upper-side and lower-side boundaries of the label. Then, the biquadratic transformation is used to restore the rectangular shape of the label. The proposed algorithm performs restoration of 3D objects in a 2D space, and it requires neither an auxiliary hardware such as 3D camera to construct 3D models nor a multi-camera to capture objects in different views. Experimental results demonstrate the effectiveness of the proposed method.

• 대한기계학회논문집A
• /
• 제25권5호
• /
• pp.841-849
• /
• 2001

The pixel information of the object was obtained sequentially and pixels were clustered to a label by the line labeling method. Feature points were determined by finding the slope for edge pixels after selecting the fixed number of edge pixels. The slope was estimated by the least square method to reduce the detection error. Once a matching point was determined by comparing the feature information of the object and the pattern, the parameters for translation, scaling and rotation were obtained by selecting the longer line of the two which passed through the matching point from left and right sides. Finally, modified Hausdorff Distance has been used to identify the similarity between the object and the given pattern. The multi-label method was developed for recognizing the patterns with more than one label, which performs the modified Hausdorff Distance twice. Experiments have been performed to verify the performance of the proposed algorithm and method for simple target image, complex target image, simple pattern, and complex pattern as well as the partially hidden object. It was proved via experiments that the proposed image matching algorithm for recognizing the object had a good performance of matching.

• International Journal of Contents
• /
• 제7권4호
• /
• pp.24-29
• /
• 2011

In this paper, we propose a method to segment the bottle label in images taken by mobile phones using multi-gradient approaches. In order to segment the label region of interest-object, the saliency map method and Hough Transformation method are first applied to the original images to obtain the candidate region. The saliency map is used to detect the most salient area based on three kinds of features (color, orientation and illumination features). The Hough Transformation is a technique to isolated features of a particular shape within an image. Therefore, we utilize it to find the left and right border of the bottle. Next, we segment the label based on the gradient information obtained from the structure tensor method and edge method. The experimental results have shown that the proposed method is able to accurately segment the labels as the first step of product label recognition system.

• Journal of Electrical Engineering and Technology
• /
• 제10권2호
• /
• pp.699-707
• /
• 2015

Although digit character recognition has got a significant improvement in recent years, it is still challenging to achieve satisfied result if the data contains an amount of distracting factors. This paper proposes a novel digit character recognition approach using a multi-layer hierarchical model, Hybrid Restricted Boltzmann Machines (HRBMs), which allows the learning architecture to be robust to background distracting factors. The insight behind the proposed model is that useful high-level features appear more frequently than distracting factors during learning, thus the high-level features can be decompose into hybrid hierarchical structures by using only small label information. In order to extract robust and compact features, a stochastic 0-1 layer is employed, which enables the model's hidden nodes to independently capture the useful character features during training. Experiments on the variations of Mixed National Institute of Standards and Technology (MNIST) dataset show that improvements of the multi-layer hierarchical model can be achieved by the proposed method. Finally, the paper shows the proposed technique which is used in a real-world application, where it is able to identify digit characters under various complex background images.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제15권6호
• /
• pp.2168-2187
• /
• 2021

The smart grid replaces the traditional power structure with information inventiveness that contributes to a new physical structure. In such a field, malicious information injection can potentially lead to extreme results. Incorrect, FDI attacks will never be identified by typical residual techniques for false data identification. Most of the work on the detection of FDI attacks is based on the linearized power system model DC and does not detect attacks from the AC model. Also, the overwhelming majority of current FDIA recognition approaches focus on FDIA, whilst significant injection location data cannot be achieved. Building on the continuous developments in deep learning, we propose a Deep Learning based Locational Detection technique to continuously recognize the specific areas of FDIA. In the development area solver gap happiness is a False Data Detector (FDD) that incorporates a Convolutional Neural Network (CNN). The FDD is established enough to catch the fake information. As a multi-label classifier, the following CNN is utilized to evaluate the irregularity and cooccurrence dependency of power flow calculations due to the possible attacks. There are no earlier statistical assumptions in the architecture proposed, as they are "model-free." It is also "cost-accommodating" since it does not alter the current FDD framework and it is only several microseconds on a household computer during the identification procedure. We have shown that ANN-MLP, SVM-RBF, and CNN can conduct locational detection under different noise and attack circumstances through broad experience in IEEE 14, 30, 57, and 118 bus systems. Moreover, the multi-name classification method used successfully improves the precision of the present identification.

• 한국정보처리학회:학술대회논문집
• /
• 한국정보처리학회 2017년도 춘계학술발표대회
• /
• pp.181-182
• /
• 2017

관성 센서 기반 행위인지는 스마트폰과 웨어러블 밴드 등의 출현으로 보다 간편한 방법으로 행위인지가 가능해졌다. 현재 대부분의 행위인지 서비스나 연구들은 단일 행위의 결론만을 도출하고 있으나, 이러한 방식은 한 행위에서 한 가지 동작밖에 취할 수 없는 경우에는 문제가 없지만 두 가지 이상의 동작이 합쳐진 경우에 어떤 행위를 최종 결론으로 도출해야 하는지에 대한 문제점을 내포한다. 따라서 본 논문에서는 이러한 문제점을 해결하기 위해 세 개의 센서 기기 (스마트폰, 스마트워치, 웨어러블 센서)를 이용한 멀티 레이블 행위인지를 제안한다. 스마트폰은 신체 전반적인 움직임 탐지를 위하여 소지위치가 정해지지 않은 비고정식 센서의 보조적인 역할을 수행한다. 스마트워치는 사용자가 주로 사용하는 손의 손목, 그리고 웨어러블 센서는 사용자의 허벅지에 부착되어 각각 상하체의 움직임을 파악한다. 이후 각 기기에서 도출된 결론에 Majority Weighted Voting 기법을 적용하여 단일 혹은 멀티 레이블의 최종 행위를 도출한다.