• 한국정보컨버전스학회:학술대회논문집
• /
• 2008.06a
• /
• pp.145-150
• /
• 2008

This thesis explores the problem of reconstructing a three-dimensional(3D) scene given a set of images or image sequences of the scene. It describes efficient methods for the 3D reconstruction of static and dynamic scenes from stereo images, stereo image sequences, and images captured from multiple viewpoints. Novel methods for image-based and volumetric modelling approaches to 3D reconstruction are presented, with an emphasis on the development of efficient algorithm which produce high quality and accurate reconstructions. For image-based 3D reconstruction a novel energy minimisation scheme, Iterated Dynamic Programming, is presented for the efficient computation of strong local minima of discontinuity preserving energyy functions. Coupled with a novel morphological decomposition method and subregioning schemes for the efficient computation of a narrowband matching cost volume. the minimisation framework is applied to solve problems in stereo matching, stereo-temporal reconstruction, motion estimation, 2D image registration and 3D image registration. This thesis establishes Iterated Dynamic Programming as an efficient and effective energy minimisation scheme suitable for computer vision problems which involve finding correspondences across images. For 3D reconstruction from multiple view images with arbitrary camera placement, a novel volumetric modelling technique, Embedded Voxel Colouring, is presented that efficiently embeds all reconstructions of a 3D scene into a single output in a single scan of the volumetric space under exact visibility. An adaptive thresholding framework is also introduced for the computation of the optimal set of thresholds to obtain high quality 3D reconstructions. This thesis establishes the Embedded Voxel Colouring framework as a fast, efficient and effective method for 3D reconstruction from multiple view images.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.28 no.7C
• /
• pp.696-704
• /
• 2003

In this paper, we propose an improved early detection of all-zero DCT coefficients for fast video encoding. From the experimental observation, it is shown that the performance of the conventional method is limited because of the imprecision sufficient condition. When the calculation of the SAD in motion estimation is simply modified, more precise sufficient condition is derived from the theoretical analysis. Based on this idea, DCT and the quantization stages are effectively skipped in the proposed algorithm with no image degradation. The simulation results show that the proposed algorithm achieves computational saving over 10% compared to the conventional method.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2015.11a
• /
• pp.203-206
• /
• 2015

본 논문은 4K UHD 입력 영상에 대한 HEVC 고속 부호화를 위하여 대부분의 상용 CPU 및 AP 에서 사용되고 있는 SIMD (Single Instruction Mutiple Data) 명령어를 사용한 고속의 정수 화소 단위 움직임 추정 방법에 대한 연구이다. 특히, IT 기기에서의 고속 동영상 부호화를 위해 기존의 SIMD 명령어를 개량하여 동일한 CPU 실행시간에 다수의 움직임 추정을 수행할 수 있는 SIMD 명령어를 사용하여 보다 같은 실행시간에 보다 넓은 영역에 대한 움직임 벡터 탐색을 수행할 수 있도록 Search Box 기법을 새로이 개발하고 이를 토대로 기존 HEVC 에서 사용되고 있는 움직임 추정 방법에 대하여 연산시간을 줄이는 동시에 화질 열화를 최소화 시킬 수 있는 방법에 대하여 논한다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2000.04a
• /
• pp.1112-1115
• /
• 2000

본 논문에서는 One-bit 변환을 기반으로 한 고속의 가변 블록 크기 움직임 예측 알고리즘을 제안한다. 제안된 방법은 블록 내의 평균값을 이용하여 8bit 화소값을 1bit로 변환한 후 움직임 예측을 수행한다. One-bit 변환을 통한 영상의 단순화는 움직임 추정의 계산적 부담을 감소시켜 빠른 탐색을 가능하게 한다. 그리고 블록 내의 움직임 정도를 미리 판별하여 이를 기반으로 한 적응적 탐색이 불필요한 탐색을 제거하고 움직임이 큰 블록에서는 정합과정을 심화시켜 보다 정확한 움직임 예측을 수행한다. 본 제안된 방식을 가지고 실험한 결과 한 프레임당 적은 수의 블록으로 고정된 크기의 블록을 가진 전역 탐색 블록 정합 알고리즘(full search block matching algorithm; FSBMA)보다 예측 에러를 적게 발생시켜 평균적으로 0.5dB 정도의 PSNR 개선을 가져왔다. 특히, 움직임이 많은 영상에서 뛰어난 효과를 나타냈다.

• Proceedings of the KSMRM Conference
• /
• 2001.11a
• /
• pp.111-111
• /
• 2001

목적： To evaluate the feasibility of the event-related functional MR study using power grip studying the hand motor system 대상 및 방법： Event-related functional MRI was performed on a 1.5T MR unit in seven norm volunteers (man=7, right-handedness=2, left-handedness=5, mean age： 25 years). A single-shot GRE-EPI sequence (TR/TE/flip angle： 1000ms/40ms/90, FOV = 240 mm matrix= 64$\times$64, slice thickness/gap = 5mm/0mm, 7 true axial slices) was used for functiona MR images. A flow-sensitive conventional gradient echo sequence (TR/TE/flip angl 50ms/4ms/60) was used for high-resolution anatomical images. To minimize the gross hea motion, neck-holders (MJ-200, USA) were used. A series of MR images were obtained in axial planes covering motor areas. To exclude motion-corrupted images, all MR images wer surveyed in a movie procedure and evaluated using the estimation of center of mass of ima signal intensities. Power grip task consisted of the powerful grip of all right fingers and hand movement ta used very fast right finger tapping at a speed of 3 per 1 second. All tasks were visual-guid by LCD projector (SHARP, Japan). Two tasks consisted of 134 phases including 7 activatio and 8 rest periods. Active stimulations were performed during 2 seconds and rest period were 15 seconds and total scan time per one task was 2 min 14 sec. Statistical maps we obtained using cross-correlation method. Reference vector was time-shifted by 4 seconds an Gaussian convolution with a FWHM of 4 seconds was applied to it. The threshold in p val for the activation sites was set to be 0.001. All mapping procedures were peformed usin homemade program an IDL (Research Systems Inc., USA) platform. We evaluated the activation patterns of the motor system of power grip compared to hand movement in t event-related functional MRI.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.17 no.7
• /
• pp.1672-1686
• /
• 2013

In this paper, we propose a hybrid super-resolution algorithm robust to cut-change. Existing single-frame based super-resolution algorithms are usually fast, but quantity of information for interpolation is limited. Although the existing multi-frame based super-resolution algorithms generally robust to this problem, the performance of algorithm strongly depends on motions of input video. Furthemore at boundary of cut, applying of the algorithm is limited. In the proposed method, we detect a define boundary of cut using cut-detection algorithm. Then we adaptively apply a single-frame based super-resolution method to detected cut. Additionally, we propose algorithms of normalizing motion vector and analyzing pattern of edge to solve various problems of existing super-resolution algorithms. The experimental results show that the proposed algorithm has better performance than other conventional interpolation methods.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.30 no.12C
• /
• pp.1256-1261
• /
• 2005

In this paper, we propose a low-power Booth multiplication which reduces the switching activities of partial products during multiplication process. Radix-4 Booth algorithm has a characteristic that produces the Booth encoded products with zero when input data have sequentially equal values (0 or 1). Therefore, partial products have higher chances of being zero when an input with a smaller effective dynamic range of two multiplication inputs is used as a multiplier data instead of a multiplicand. The proposed multiplier divides a multiplication expression into several multiplication expressions with smaller bits than those of an original input data, and each multiplication is computed independently for the Booth encoding. Finally, the results of each multiplication are added. This means that the proposed multiplier has a higher chance to have zero encoded products so that we can implement a low power multiplier with the smaller switching activity. Implementation results show the proposed multiplier can save maximally about $20\%$ power dissipation than a previous Booth multiplier.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2022.06a
• /
• pp.205-208
• /
• 2022

Multi-exposure high dynamic range (HDR) image reconstruction, the task of reconstructing an HDR image from multiple low dynamic range (LDR) images in a dynamic scene, often produces ghosting artifacts caused by camera motion and moving objects and also cannot deal with washed-out regions due to over or under-exposures. While there has been many deep-learning-based methods with motion estimation to alleviate these problems, they still have limitations for severely moving scenes. They also require large parameter counts, especially in the case of state-of-the-art methods that employ attention modules. To address these issues, we propose a frequency domain approach based on the idea that the transform domain coefficients inherently involve the global information from whole image pixels to cope with large motions. Specifically we adopt Residual Fast Fourier Transform (RFFT) blocks, which allows for global interactions of pixels. Moreover, we also employ Depthwise Overparametrized convolution (DO-conv) blocks, a convolution in which each input channel is convolved with its own 2D kernel, for faster convergence and performance gains. We call this LFFNet (Lightweight Frequency Fusion Network), and experiments on the benchmarks show reduced ghosting artifacts and improved performance up to 0.6dB tonemapped PSNR compared to recent state-of-the-art methods. Our architecture also requires fewer parameters and converges faster in training.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.4 no.3
• /
• pp.126-133
• /
• 2003

This paper is related to a morphological segmentation method for extracting the moving object in video sequence using global motion compensation and two-dimensional spatio-temporal entropic thresholding. First, global motion compensation is performed with camera panning vector estimated in the hierarchical pyramid structure constructed by wavelet transform. Secondly, the regions with high possibility to include the moving object between two consecutive frames are extracted block by block from the global motion compensated image using two-dimensional spatio-temporal entropic thresholding. Afterwards, the LUT classifying each block into one among changed block, uncertain block, stationary block according to the results classified by two-dimensional spatio-temporal entropic thresholding is made out. Next, by adaptively selecting the initial search layer and the search range referring to the LUT, the proposed HBMA can effectively carry out fast motion estimation and extract object-included region in the hierarchical pyramid structure. Finally, after we define the thresholded gradient image in the object-included region, and apply the morphological segmentation method to the object-included region pixel by pixel and extract the moving object included in video sequence. As shown in the results of computer simulation, the proposed method provides relatively good segmentation results for moving object and specially comes up with reasonable segmentation results in the edge areas with lower contrast.

• Journal of Broadcast Engineering
• /
• v.2 no.2
• /
• pp.86-93
• /
• 1997

Fast growth of multimedia applications requires new functions for video data processing. such as obj;cted-based video representation and manipulation. which are not supported by 11PEG-l and 11PEG-2. To support these requirements. 11PEG-4 video coding allows users to manipulate every video object easily by decomposing a scene into several video objects and coding each of them independently. However. the large amount of computations and flexible structure of 11PEG-4 video CODEC make it difficult to be implemented by either the general purpose DSP or a dedicated VLSI. In this paper, we propose a hardware architecture using a hybrid of a high performance programmable DSP and an application specific IC to implement a flexible 11PEG-4 video codec requiring the large amount of computations. The application specific IC has the functions of motion estimation and compensation.