Chi-Wah Wong

Efficient intra-prediction algorithm in H.264

In the upcoming H.264, intra-prediction with block sizes of 4/spl times/4 and 16x16 is used to compress I-frame. However, there are 9 or 4 candidate modes for the 4x4 or 16x16 intra-prediction. The full search algorithm to choose the modes is computationally expensive. In this paper, we suggest an efficient intra-prediction (EIP) algorithm based on early termination, selective computation of highly probable modes, and partial computation of the cost function. The proposed EIP reduces the complexity considerably while maintaining similar PSNR and bit rate.

Automatic white balancing using standard deviation of RGB components

A good automatic white balancing algorithm is an essential image pre-processing component in consumer digital still cameras, and it can greatly improve the final image quality. In this paper, an automatic white balancing algorithm based on the standard deviation of RGB components of the pre-captured image is proposed. A light source model for evaluation of an automatic white balancing is also described. The simulation results show that the proposed algorithm can improve the final image quality of the captured image.

Perceptual rate control for low-delay video communications

The existing rate control schemes in the literature evaluate their video quality performance in terms of traditional PSNR. Actually, this evaluation may not be good because this measurement is not the same as that judged by human. This work defines a perceptual quality measure based on the human vision system. The modified rate control includes the perceptual consideration. The results show that the perceptual rate control achieves average perceptual quality gain over TMN8.

Efficient intra-prediction mode selection for 4/spl times/4 blocks in H.264

In the upcoming H.264, intra-prediction for 4/spl times/4 and 16/spl times/16 blocks is used to compress I-frame. However, we need to choose one of the 9 prediction modes for each 4/spl times/4 block. The full search (FS) algorithm to select the mode is computationally expensive. In this paper, an efficient intra-prediction mode selection (EIPMS) method is proposed based on partial computation of the cost function, selective computation of highly probable modes, early termination, and variable threshold setting. The proposed EIPMS can reduce the complexity considerably while maintaining similar PSNR and bit rate.

Novel H.26X optimal rate control for low-delay communications

The existing rate control schemes in the literature calculates the quantization parameters of macro-blocks (MB) based on the quadratic/log rate models. Actually, this model may not be good because this model does not estimate well at some particular bit rates. This work investigates the relationships of bit rate and distortion with the MB quantization step size and standard deviation of MB prediction errors. We find that the linear rate and linear distortion models are established. Then we present the proposed rate control schemes based on these models. The proposed algorithm introduces a low quantization overhead, low computation complexity to calculate the optimal quantization step size, independence of the distortion factors and low memory usage compared with TMN8 rate control. The experimental results suggests that our scheme can achieve PSNR gain over TMN8.

Piecewise linear model for real-time rate control [video coding applications]

Most existing rate control schemes in the literature evaluate quantization parameters of macro-blocks (MB) based on models which estimate well at low bit rates (e.g. quadratic model). These models are not flexible because this model does not estimate well for different situations with different rates. This work investigates the relationships of the rate and the distortion of a residue MB with two parameters - MB quantization step size and standard deviation of MB prediction errors. A piecewise linear rate model and linear distortion model are proposed. Then, we present the proposed rate control scheme based on these models. The experimental results suggest that our scheme achieves PSNR gain over TMN8 and Lis scheme.

Automatic white balancing using luminance component and standard deviation of RGB components [image preprocessing]

Automatic white balancing is an essential image preprocessing component in consumer digital still cameras, and it can greatly improve the final image quality of the captured image. In this paper, a novel automatic white balancing algorithm based on both the luminance component and standard deviation of RGB components of the pre-captured image is proposed. A light source model for evaluation of an automatic white balancing is also described. The simulation results indicate that the proposed algorithm can improve the final image quality of the captured image.

Real-Time rate control via variable frame rate and quantization parameters

Most of existing rate control schemes in the literature focus on the bit allocation under the assumption of constant frame rate. As a result, the distortion of consecutive frames varies in a great extent, which may result in low temporal quality (e.g. flickering effect). In this work, we design rate control with two control parameters (i.e. quantization factor and frame rate) to achieve better tradeoff between spatial quality and temporal quality in low complexity. Our optimization gives two suggested quantization step sizes. One is used to change the frame rate in order to allocate how many bits used in the current frame. Another one is used to change the quantization factor in order to quantize MBs to achieve target bit rates. The experimental results suggest that our scheme can achieve more consistent quality while keeping high spatial quality.

Advanced Macro-block Entropy Coding in H.264

Existing video encoders, such as MPEG4, H.263 and H.264, adopt variable length coding (VLC) as entropy coding method, focusing on residue data coding. In low bit-rate video coding, larger quantization parameters are used to give a smaller number of bits spent on residue data. As a result, the header overhead is a dominant factor of yielding the overall bit rates. In this paper, macro-block (MB) header behavior is investigated. It is found that the relative percentage of the number of bits of MB header increases with quantization parameter (QP) and header correlation between neighboring MBs is high. Based on these MB header properties, we propose an advanced coding algorithm for entropy coding of MB header. The experimental results suggest that our proposed algorithm achieves lower total number of encoded bits over JM10.2, up to 10% bit reduction, with the same PSNR quality. At the same bit rates, our algorithm has about 0.3-0.4 dB PSNR gain.

Fast motion vector re-estimation for arbitrary video downsizing using spatial-variant filter

To downsize a compressed video to a lower resolution, the straightforward method is to decompress the compressed video, downsample the video in spatial domain and then re-compress the downsampled video. This process can be computationally expensive unless fast motion vector re-estimation methods are used. In this paper, the use of a spatial-variant filter (SVF) as a fast motion vector re-estimation method for arbitrary downsizing factor is investigated. Simulation results suggest that the SVF can improve the quality of the existing fast algorithms in terms of PSNR. With a little bit of local search, its PSNR can be close to the full search