Hoonjae Lee

Spatial error concealment for H.264 using sequential directional interpolation

Error concealment at the decoder restores erroneous macroblocks (MBs) caused by channel errors. In this paper, we propose a novel spatial error concealment algorithm based on prediction modes of intra-blocks which are included in a H.264-coded stream and highly correlated to the direction of local edge within the block. The key contribution is to sequentially interpolate each pixel in a lost MB by utilizing edge directions and strengths efficiently estimated from the neighboring blocks, preserving local edge continuity for more visually acceptable images. The proposed scheme is simple to implement and more reliably recover high-detailed content in corrupted MBs. The experimental results shows the proposed method achieves reduction in speed by 14%~39% as compared to existing method, and outperforms them in PSNR by 0.5~1 dB as well as in subjective visual evaluation.

A VBR Video Encoding for Locally Consistent Picture Quality With Small Buffering Delay Under Limited Bandwidth

During consecutive large picture information changes under limited bandwidth, the use of constant bit rate (CBR) encoding for high definition television (HDTV) broadcasting systems often causes serious picture quality degradation due to the tight constraint on small buffering delay, whereas a variable allocation of bits depending on the picture information changes results in large buffering delay. In this paper, we propose a frame-layer variable bit rate (VBR) video encoding method for maintaining locally consistent picture quality of an encoded video with small buffering delay, which can be used for a video on demand (VOD) system which streams an encoded video file to a client device under limited bandwidth or a real-time streaming application allowing a small delay. An objective function defined as a sum of the local distortion variation of the encoded pictures and the underflow levels at the decoder buffer is minimized with respect to the discrete frame-layer quantization step sizes subject to a constraint on the target number of bits within a temporal sliding window. Then, the constrained discrete minimization problem is converted to an unconstrained continuous minimization problem by introducing a Lagrange multiplier and is solved by applying the first-order necessary condition for local minima with respect to the quantization step sizes. Experimental results demonstrate that the proposed method provides a lower local standard deviation of PSNR comparing with the recent VBR method by 29.2% on average with small buffering delay.

Efficient Viewer-Centric Depth Adjustment Based on Virtual Fronto-Parallel Planar Projection in Stereo 3D Images

This paper presents an efficient method for adjusting the 3D depth of an object including as much as a whole scene in stereo 3D images by utilizing a virtual fronto-parallel planar projection in the 3D space perceived by a viewer. The proposed method just needs to establish object correspondence instead of the accurate estimation of the disparity field or point correspondence. We simulate the depth adjustment of a 3D point perceived by a viewer through a corresponding pair of points on the stereo 3D images by moving the virtual fronto-parallel plane on which the left and right points are projected. We show that the resulting transformation of image coordinates of the points can be simply expressed by three values of a scale factor and two translations that depend on one parameter for the depth adjustment. The experimental results demonstrate the feasibility of the proposed approach that yields less visual fatigue and smaller 3D shape distortion than the conventional parallax adjustment method. The overall procedure can be efficiently applied to each frame of a stereo video without causing any artifact.

Fast thumbnail generation in integer DCT domain for H.264/AVC

In this paper, we propose a simple and efficient method for generating a thumbnail or 1/4 reduced-size image in integer Discrete Cosine Transform (DCT) domain for H.264/AVC video streams. To accomplish this, we first obtain the integer DCT coefficients of a prediction residual using a newly-defined re-scaling operation of inverse quantization, and then perform a transform domain intra prediction in integer DCT domain through simplified matrix multiplication by using only shift operation and integer addition. Finally, we generate a thumbnail directly from a H.264/AVC intra frame. Experimental results demonstrate that the proposed algorithm can reduce the total computational complexity about 16% on average, while maintaining almost the same image quality as compared to the existing method.

An Optimized Backlight Local Dimming Algorithm for Edge-Lit LED Backlight LCDs

This paper presents an efficient optimized backlight local dimming method that minimizes power consumption under a given allowable distortion for liquid crystal displays with edge-lit light-emitting diode backlight. To reduce image quality fluctuation, an inequality constraint optimization problem is formulated using a given allowable distortion and the steepest descent method is used to solve the optimization problem. The experimental results show that the proposed method has better performance than previous methods at the same average power consumption in the unnoticeable level of the difference between the displayed images before and after dimming. Also, the proposed method reduces the risk of large performance degradation and reduces computation time by three times compared with the existing optimization based methods.

Adaptive parallax control based on 3D scene change detection

This paper presents an adaptive parallax control method based on 3D scene change detection for stereoscopic video sequences. For each pair of left and right frames, the maximum disparity corresponding to the closest object is estimated by using a simple block matching algorithm. The 3D scene change is detected if there exists an abrupt change in the values of either the maximum disparity or color histogram between successive pair of frames. Then, if the maximum disparity is excessive at a pair of frames where the scene change has occurred, the horizontal parallax is controlled by reducing the excessive maximum disparity that could induce the visual discomfort. The experimental results demonstrate the feasibility of the proposed approach.

Efficient transform domain transcoding: intra frame of H.264/AVC to JPEG

In this paper, we propose an efficient transform domain method for transcoding an intra frame of H.264/AVC to a JPEG image. First, we show how to obtain the transform domain predictions for various intra modes of H.264/AVC through simplified matrix multiplication. Then, we describe a transform domain operation corresponding to the rounding operation which is performed after computing the weighted average during spatial domain intra prediction. Finally, we convert 4 × 4 integer transform blocks to 8 × 8 DCT blocks by defining a transform kernel matrix based on the factorization of transform matrices. Experimental results demonstrate that the proposed algorithm can reduce the total computational complexity about 16% on average as compared with the existing methods, and outperforms them in PSNR by 1.5~2.5dB.

Frame layer CBR encoding to reduce large picture quality fluctuation

Constant bit-rate (CBR) encoding methods used for the real-time digital video applications such as terrestrial high definition television (HDTV) broadcasting often produce large picture quality fluctuation due to constraints on buffer size and buffering delay. In this paper, we propose a simple frame-layer CBR encoding method which reduces large picture quality fluctuation of an encoded video under constant bit-rate and limited buffer size. We first define an intermediate encoder buffer level to be lower than the half of the maximum encoder buffer level. Then, the encoder buffer level is kept from being close to the maximum (or minimum) buffer level during encoding by slightly decreasing (or increasing) the number of bits estimated to encode each frame if the resulting buffer level becomes higher (or lower) than the intermediate buffer level. The experimental results show that the proposed CBR encoding method reduces the average local PSNR variation by 20.6% compared with previous CBR encoding method on average under constant bit-rate and limited buffer size.

Image Enhancement for Reducing LCD Backlight Power Based on Human Visual Characteristics

One common way to reduce power consumption of display devices is to reduce the liquid crystal displays (LCD) backlight intensity since the main light source of LCD consumes 20~50% of the total system power. In this paper, we propose a novel method of reducing LCD backlight power which maximizes the power reduction ratio and minimizes the visual image quality degradation based on the human visual system (HVS). The proposed method uses histogram to minimize the number of clipped pixels and keep the perceptual quality and the characteristics of image. In contrast to previous approach, the proposed method uses no look-up table and preserves the hue values and the perceptual white brightness. Experimental results show that 41.35% of backlight power is reduced on average without severe perceptual distortion.

Rapid hybrid encoding for high-resolution whole-brain fluid-attenuated imaging

Single‐slab three‐dimensional (3D) turbo spin‐echo (TSE) imaging combined with inversion recovery (IR), which employs short, spatially non‐selective refocusing pulses and signal prescription based variable refocusing flip angles (VFA) to increase imaging efficiency, was recently introduced to produce fluid‐attenuated brain images for lesion detection. Despite the advantages, the imaging efficiency in this approach still remains limited because a substantially long time of inversion is needed to selectively suppress the signal intensity of cerebrospinal fluid (CSF) while fully recovering that of brain tissues. The purpose of this work is to develop a novel, rapid hybrid encoding method for highly efficient whole‐brain fluid‐attenuated imaging. In each time of repetition, volumetric data are continuously encoded using the hybrid modular acquisition in a sequential fashion even during IR signal transition, wherein reversed fast imaging with steady‐state free precession (PSIF) is employed to encode intermediate‐to‐high spatial frequency signals prior to CSF nulling, while VFA–TSE is used to collect low‐to‐intermediate spatial frequency signals afterwards. Gradient‐induced spin de‐phasing between a pair of neighboring radio‐frequency (RF) pulses in both PSIF and TSE modules is kept identical to avoid the occurrence of multiple echoes in a single acquisition window. Additionally, a two‐step, alternate RF phase‐cycling scheme is employed in the low spatial frequency region to eliminate free induction decay induced edge artifacts. Numerical simulations of the Bloch equations were performed to evaluate signal evolution of brain tissues along the echo train while optimizing imaging parameters. In vivo studies demonstrate that the proposed technique produces high‐resolution isotropic fluid‐attenuated whole‐brain images in a clinically acceptable imaging time with substantially high signal‐to‐noise ratio for white matter while retaining lesion conspicuity. Copyright