• 한국정보처리학회:학술대회논문집
• /
• 한국정보처리학회 2019년도 춘계학술발표대회
• /
• pp.515-518
• /
• 2019

Semantic segmentation has remained as a challenging problem in the field of computer vision. Given the immense power of Convolution Neural Network (CNN) models, many complex problems have been solved in computer vision. Semantic segmentation is the challenge of classifying several pixels of an image into one category. With the help of convolution neural networks, we have witnessed prolific results over the time. We propose a convolutional neural network model which uses Fully CNN with deconvolutional pixel layers. The goal is to create a hierarchy of features while the fully convolutional model does the primary learning and later deconvolutional model visually segments the target image. The proposed approach creates a direct link among the several adjacent pixels in the resulting feature maps. It also preserves the spatial features such as corners and edges in images and hence adding more accuracy to the resulting outputs. We test our algorithm on Karlsruhe Institute of Technology and Toyota Technologies Institute (KITTI) street view data set. Our method achieves an mIoU accuracy of 92.04 %.

• Journal of Information Processing Systems
• /
• 제15권5호
• /
• pp.1082-1095
• /
• 2019

Camera calibration is an important part of machine vision and close-range photogrammetry. Since current calibration methods fail to obtain ideal internal and external camera parameters with limited computing resources on mobile terminals efficiently, this paper proposes an improved fast camera calibration method for mobile terminals. Based on traditional camera calibration method, the new method introduces two-order radial distortion and tangential distortion models to establish the camera model with nonlinear distortion items. Meanwhile, the nonlinear least square L-M algorithm is used to optimize parameters iteration, the new method can quickly obtain high-precise internal and external camera parameters. The experimental results show that the new method improves the efficiency and precision of camera calibration. Terminals simulation experiment on PC indicates that the time consuming of parameter iteration reduced from 0.220 seconds to 0.063 seconds (0.234 seconds on mobile terminals) and the average reprojection error reduced from 0.25 pixel to 0.15 pixel. Therefore, the new method is an ideal mobile terminals camera calibration method which can expand the application range of 3D reconstruction and close-range photogrammetry technology on mobile terminals.

• 대한임베디드공학회논문지
• /
• 제14권3호
• /
• pp.157-163
• /
• 2019

As the drone industry has grown greatly in recent years, drones are being used or developed in many industrial fields such as image shooting, pesticide application, delivery service, food delivery etc. In this paper, therefore, we developed a program that takes a user's desired area at a certain height using a camera-equipped drone and obtains the area of the zone the user wants through image processing. The first user selects an area or a path. Afterwards, the drone flies and takes pictures, and then measures the user's needs. A digital image taken at a constant height and with the same resolution is composed of pixels, the area can be calculated easily if we know the number of pixels in the zone the user wants. Particularly, it is easy to calculate the area of various shaped zones, not terrain shapes such as triangles and squares. In addition, the total area of specific places of the entire zone can be calculated. With the program of this paper, anyone can easily calculate the area of the place the user wants using a drone rather than calculating the area through difficult formulas or specialized equipment.

• 대한전자공학회:학술대회논문집
• /
• 대한전자공학회 2000년도 추계종합학술대회 논문집(3)
• /
• pp.153-156
• /
• 2000

A novel silicon retina chip based on the information processing in the vertebrate retina was designed. The chip has a novel wiring structure in which all pixels are connected through the channel of MOS transistors, which simplifies a wiring structure compared with conventional resistive networks. The proposed structure minimizes the pixel area and certainly increases a fill factor since each pixel consists of only two photodiodes and three MOS transistors. It also enables the chip to operate over a wide range of light intensity by adjusting its conductance with the gate voltage. Simulation results with SPICE showed that the chip could extract the edge of input images successfully.

• 한국정보전자통신기술학회논문지
• /
• 제11권5호
• /
• pp.601-607
• /
• 2018

Median filtering is a nonlinear method which is known to be effective in removing impulse noise while preserving local image structure relatively well. However, it could still suffer the smearing phenomena of edges and fine details into neighbors due to undesirable influence from the pixels whose values are far off from the true value of the pixel at hand. This drawback mainly comes from the fact that median filters typically employ a regular shape window for collecting the pixels used in the filtering operation. In this paper, we propose a median filtering method which employs an irregular shape filter window in collecting neighboring pixels around the pixel to be denoised. By employing an irregular shape window, we can achieve good noise suppression while preserving image details. Experimental results have shown that our approach is superior to regular window-based methods.

• 대한원격탐사학회:학술대회논문집
• /
• 대한원격탐사학회 2002년도 Proceedings of International Symposium on Remote Sensing
• /
• pp.536-541
• /
• 2002

This paper proposes an approach of automatically calculating the center location of fiducial marks up to sub-pixel accuracy with model-free condition. The conceptual model, composed of two assumptions, about general geometric property of fiducial area and fiducial mark is established. The proposed approach is primarily based on the strategy of calculating the center of symmetry and is composed of three steps of processing: (a) determining horizontal center of fiducial area, (b) locating center of a fiducial mark, and, (c) finally calculating the exact center up to sub-pixel accuracy. Evaluation with respect to RMK style marks and RC style ones shows that the proposed approach can be evaluated as robust one. However there are several images which can't be analyzed by the proposed approach.

• 대한전자공학회논문지SP
• /
• 제46권6호
• /
• pp.68-77
• /
• 2009

본 논문에서는 저해상도 유추 보간법(interpolation algorithm derived from low resolution: ILR)을 이용한 공간적 디인터레이싱 기법을 제안한다. 일반적인 에지 기반 디인터레이싱 방법들은 화소 단위의 상관도를 이용하기 때문에 잡음과 밝기의 변화에 민감하다는 단점이 있다. 또 보간의 성능을 좌우하는 정확한 에지 방향을 판단함에 있어서 만족스럽지 못한 성능을 보인다. 이러한 한계를 극복하기 위해 본 논문에서는 먼저 보간 하고자 하는 픽셀 주위의 저해상도 블록의 특성을 먼저 탐구하고 이를 고해상도 블록에 적용함으로써 missing pixel 을 보간 하는 방법을 제안한다. 실험결과 제안하는 방법은 기존의 화소 단위의 에지 기반 디인터레이싱 방법보다 PSNR로 대표되는 객관적 성능과 주관적 화질 측면에서 우수한 결과를 나타내는 것을 볼 수 있었다.

• 한국정보처리학회논문지
• /
• 제7권8호
• /
• pp.2565-2574
• /
• 2000

오차확산법은 연속계조 영상을 중간조 영상으로 생성시 우수한 재현성을 보인다. 그러나 표시오차의 전력스펙트럼 분석에서 경계정보의 재현성이 다소 떨어지는 특성을 보인다. 이를 개선하기 위해 원영상의 현재화소와 로컬 평균간의 차이정보를 이용하는 경계강조 오차확산법을 제안한다. 제안한 기법은 원영상이 현재화소와 로컬 평균과의 차이정보 및 이를 활용하는 필터의 가중치 함수로 구성된다. 첫째, 원영상의 차이정보는 현재 화소와 이의 인접화소(5x5)의 로컬 평균과의 차이이다. 둘째, 필터의 가중치 함수는 차이정보의 크기를 포함하는 함수와 이의 부호로 구성된다. 제안한 기법을 적용한 중간조 영상은 경계가 강조되어 시각적으로 선명한 결과를 보인다. 환상 평균 전력 스펙트럼 밀도를 이용한 표시오차, 경계상관도 및 로컬 평균 일치도의 평가함수로 제안한 경계강조 오차확산법과 기존의 경계강조 오차확산법의 특성을 비교한다.

• Journal of information and communication convergence engineering
• /
• 제8권6호
• /
• pp.716-720
• /
• 2010

This paper proposes an effective fingerprint identification system with hardware block for thinning stage processing of a verification algorithm based on minutiae with 39% occupation of 32-bit RISC microprocessor cycle. Each step of a fingerprint algorithm is analyzed based on FPGA and ARMulator. This paper designs an effective hardware scheme for thinning stage processing using the Verilog-HDL in $160{\times}192$ pixel array. The ZS algorithm is applied for a thinning stage. The logic is also synthesized in $0.35{\mu}m$ 4-metal CMOS process. The layout is performed based on an auto placement-routing and post-simulation is performed in logic level. The result is compared with a conventional one.

• IEIE Transactions on Smart Processing and Computing
• /
• 제4권4호
• /
• pp.265-271
• /
• 2015

This study proposes an algorithm for recognizing apple trees in images and detecting apples to measure the number of apples on the trees. The proposed algorithm explores whether there are apple trees or not based on the number of image block-unit edges, and then it detects apple areas. In order to extract colors appropriate for apple areas, the CIE $L^*a^*b^*$ color space is used. In order to extract apple characteristics strong against illumination changes, modified census transform (MCT) is used. Then, using the AdaBoost learning algorithm, characteristics data on the apples are learned and generated. With the generated data, the detection of apple areas is made. The proposed algorithm has a higher detection rate than existing pixel-based image processing algorithms and minimizes false detection.