• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2020.07a
• /
• pp.232-235
• /
• 2020

최근 딥러닝을 이용한 자동차 번호판 인식 알고리즘에 있어서 인조 번호판을 생성하여 데이터 수집과 라벨링 작업 시간을 줄이기 위한 연구가 진행되고 있다. 하지만 인조 번호판의 특성상 정면의 이미지로 구성되어 있기 때문에 자동차의 정면에서 촬영된 번호판의 인식률은 높지만 측면에서 촬영된 번호판의 경우 인식률이 낮아진다. 본 논문에서는 다양한 카메라 설치 위치에 따른 다각도로 촬영된 번호판 영상의 인식률을 보완하기 위해 이미지를 3차원으로 회전하여 데이터를 생성하는 인조 번호판 생성기 프로그램을 개발하였다. 3차원 회전을 하였을 때 번호판 인식 성능을 비교하기 위해 기존 방식으로 생성한 번호판과 제안 방식으로 생성한 번호판 각 600,000장씩 생성하여 총 1,200,000장을 생성하였으며, 데이터의 비율에 따라 10가지의 학습 데이터 셋을 구성하였다. 인조 번호판 데이터의 학습 결과를 평가하기 위해 실제 번호판 이미지 1789장으로 테스트 셋을 구성하였고, 기존의 인조 번호판 생성 방식과 인식 정확도를 비교 분석하였다.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• fall
• /
• pp.74-77
• /
• 2021

최근 딥러닝을 이용하여 객체 인식 학습을 위한 데이터셋을 구축하는데 있어 시간과 인력을 단축하기 위해 인조 데이터를 생성하는 연구가 진행되고 있다. 하지만 실제 환경과 관계없이 임의의 배경에 배치되어 구축된 데이터셋으로 학습된 네트워크를 실제 환경으로 구성된 데이터셋으로 테스트할 경우 인식률이 저조하다. 이에 본 논문에서는 실제 배경 이미지에 객체 이미지를 합성하고, 다양성을 위해 3차원으로 회전하여 증강하는 인조 데이터셋 생성 시스템을 제안한다. 제안된 방법으로 구축된 인조 데이터셋으로 학습한 네트워크와 실제 데이터셋으로 학습된 네트워크의 인식률을 비교한 결과, 인조 데이터셋의 성능이 실제 데이터셋의 성능보다 2% 낮았지만, 인조 데이터셋을 구축하는 시간이 실제 데이터셋을 구축하는 시간보다 약 11배 빨라 시간적으로 효율적인 데이터셋 구축 시스템임을 증명하였다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2022.11a
• /
• pp.390-392
• /
• 2022

기계학습 분야에서 모델을 학습시키려면 많은 양의 데이터가 필요하다. 최근에는 컴퓨터 비전 분야에서 데이터가 적은 환경에서 모델을 학습하는 다양한 방법들이 소개되고 있다. 하지만 대부분의 방법을 사용하기 위해서는 어느 정도 최소한의 학습 데이터가 필요하기 때문에 극심하게 데이터가 부족한 환경에서는 사용하기 어렵다. 본 논문에서는 컴퓨터 비전 분야에서 기계학습을 사용할 때 극심하게 데이터가 부족한 환경에서 시뮬레이션 도구를 활용한 인조 데이터 생성 방법을 제안한다. 실험 결과를 통해 시뮬레이션 도구를 활용하여 생성한 인조 데이터로 학습한 모델이 실제 데이터만을 학습한 모델을 대체할 수 있음을 확인하였고, F-1 점수와 정확도가 향상함을 실험적으로 확인하였다.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.25 no.6
• /
• pp.774-784
• /
• 2021

The object detection and recognition process is a very important task in the field of computer vision, and related research is actively being conducted. However, in the actual object recognition process, the recognition accuracy is often degraded due to the resolution mismatch between the training image data and the test image data. To solve this problem, in this paper, we designed and developed an integrated object recognition and super-resolution framework by proposing an image super-resolution technique to improve object recognition accuracy. In detail, 11,231 license plate training images were built by ourselves through web-crawling and artificial-data-generation, and the image super-resolution artificial neural network was trained by defining an objective function to be robust to the image flip. To verify the performance of the proposed algorithm, we experimented with the trained image super-resolution and recognition on 1,999 test images, and it was confirmed that the proposed super-resolution technique has the effect of improving the accuracy of character recognition.

• Journal of Broadcast Engineering
• /
• v.27 no.2
• /
• pp.185-197
• /
• 2022

Recently, deep learning networks with high performance for object recognition are emerging. In the case of object recognition using deep learning, it is important to build a training data set to improve performance. To build a data set, we need to collect and label the images. This process requires a lot of time and manpower. For this reason, open data sets are used. However, there are objects that do not have large open data sets. One of them is data required for license plate detection and recognition. Therefore, in this paper, we propose an artificial license plate generator system that can create large data sets by minimizing images. In addition, the detection rate according to the artificial license plate arrangement structure was analyzed. As a result of the analysis, the best layout structure was FVC_III and B, and the most suitable network was D2Det. Although the artificial data set performance was 2-3% lower than that of the actual data set, the time to build the artificial data was about 11 times faster than the time to build the actual data set, proving that it is a time-efficient data set building system.

• Journal of Broadcast Engineering
• /
• v.26 no.4
• /
• pp.453-462
• /
• 2021

A lot of license plate data is required for car number recognition. License plate data needs to be balanced from past license plates to the latest license plates. However, it is difficult to obtain data from the actual past license plate to the latest ones. In order to solve this problem, a license plate recognition study through deep learning is being conducted by creating a synthetic license plates. Since the synthetic data have differences from real data, and various data augmentation techniques are used to solve these problems. Existing data augmentation simply used methods such as brightness, rotation, affine transformation, blur, and noise. In this paper, we apply a style transformation method that transforms synthetic data into real-world data styles with data augmentation methods. In addition, real license plate data are noisy when it is captured from a distance and under the dark environment. If we simply recognize characters with input data, chances of misrecognition are high. To improve character recognition, in this paper, we applied the DeblurGANv2 method as a quality improvement method for character recognition, increasing the accuracy of license plate recognition. The method of deep learning for license plate detection and license plate number recognition used YOLO-V5. To determine the performance of the synthetic license plate data, we construct a test set by collecting our own secured license plates. License plate detection without style conversion recorded 0.614 mAP. As a result of applying the style transformation, we confirm that the license plate detection performance was improved by recording 0.679mAP. In addition, the successul detection rate without image enhancement was 0.872, and the detection rate was 0.915 after image enhancement, confirming that the performance improved.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.45 no.1
• /
• pp.55-63
• /
• 2008

Ridge orientations of fingerprint image are crucial informations in many parts of fingerprint recognition such as enhancement, matching and classification. Therefore it is essential to extract the ridge orientations of image accurately because it directly affects the performance of the system. The two main properties of ridge orientation are 1) global characteristic(gradual change in whole part of fingerprint) and 2) local characteristic(abrupt change around core and delta points). When we only consider the local characteristic, estimated ridge orientations are well around singular points but not robust to noise. When the global characteristic is only considered, to estimate ridge orientation is robust to noise but cannot represent the orientation around singular points. In this paper, we propose a novel method for estimating ridge orientation which represents local characteristic specifically as well as be robust to noise. We reduce the noise caused by scar using iterative outlier rejection. We apply adaptive measurement resolution in each fingerprint area to estimate the ridge orientation around singular points accurately. We evaluate the performance of proposed method using synthetic fingerprint and FVC 2002 DB. We compare the accuracy of ridge orientation. The performance of fingerprint authentication system is evaluated using FVC 2002 DB.