고효율 무손실 영상 압축을 위한 다단계 공간적 예측과 엔트로피 부호화

Abstract

High Efficiency Video Coding (HEVC) is a new standard for video compression developed jointly Video Coding Experts Group (VCEG) and Moving Picture Experts Group (MPEG). HEVC achieves significant compression performance relative to previous standard such as H.264/Advanced Video Coding (AVC), with bit-rate reductions in the range of 50% for the same subjective quality. After the first standardization of HEVC was finalized in early 2013, several extensions for HEVC were presented under active development, such as HEVC Range Extension (RExt), scalable video coding (SVC), and 3D video coding. In case of RExt, it has a goal to support extended color formats, higher bit depths, and lossless compression. Among these, lossless compression is widely exploited in various fields such as medical image, remote sensing, and image analysis. And demands of lossless compression are increased speedily. Various coding techniques are included in HEVC and significant coding efficiency is achieved through these techniques. However, HEVC is designed for lossy compression, and it is not ideal for lossless compression. Therefore, it is difficult to expect good compression performance for lossless compression. In HEVC, the basic concept for lossless compression bypasses transform and quantization, as well as their inverse processes. In-loop filtering is also bypassed, leaving only two remaining parts: prediction and entropy coding. Therefore, for achieving good compression performance in lossless coding, it is important to improve the conventional prediction and entropy coding. In the case of intra prediction, the basic structure is block-based coding. In the case of block-based coding, pixels that are located farther away are exploited for prediction. Because of the block-based coding, the accuracy of prediction may be reduced. In order to solve this problem, pixel-based coding tool is required. However, the pixel-based prediction for lossless coding may cause a problem of harmony with the block-based structure for lossy coding. Therefore, it is important to improve the compression performance with the block-based coding structure. In entropy coding, the allocation of proper codewords according to the coefficient level is important. However, the current design for entropy coding in HEVC focuses on the efficient coding of the coefficient that occurs in the lossy coding environment. When wasteful bits occur due to the allocation of an improper codewords to the coefficient in lossless coding environment, the total compression performance may decrease. Therefore, it is important to accurately predict the coefficient and efficiently allocate the codewords. To overcome the limitations of the conventional coding structure, this dissertation focuses on improvement of intra prediction and entropy coding for lossless compression. For improvement of lossless intra prediction, multi-stage intra prediction is proposed for lossless coding. In the first stage, conventional block-based intra prediction is applied based on the intra-prediction mode. In the second stage, pixel-based prediction is applied to the residuals obtained from the intra prediction of the previous stage. In the pixel-based prediction, it is possible to exploit pixels that are closer than the block-based prediction, since all the pixels can be accurately reconstructed without any data loss. In the second stage, pixel-based prediction is achieved by applying residual differential pulse code modulation (RDPCM). In comparison to HEVC, the accuracy of intra prediction can increase by the pixel-based prediction. Although RDPCM is based on pixel-based coding, it is also possible to maintain the block-based coding structure through the RDPCM. In the final stage of the proposed intra prediction, an additional prediction for the residuals is applied by exploiting the linear variation between adjacent pixels. This leads to an additional improvement in the accuracy of the prediction. Thus, the proposed intra prediction aims to improve compression performance in lossless coding by improving the accuracy of the prediction by utilizing multi-stage prediction. In entropy coding, efficient coefficient level coding method for the residuals of lossless coding is proposed. In order to efficiently encode the level information of coefficient, several parameters related to the coefficient coding are modified based on characteristic of residuals in lossless coding environment. For example, the coefficient scanning pattern and the rice parameter are the candidates for the parameters. Above all, it is important to predict the following non-zero coefficient. In the proposed method, a binarization table for the allocation of codewords to the non-zero coefficient is selected based on the threshold condition obtained by applying the average of the previous non-zero coefficients. In the proposed method, the previous non-zero coefficients influence the prediction of the next non-zero coefficient and the selection of binarization table. This dissertation achieves good compression performance by proposing coefficient coding method to consider the characteristic of coefficient in lossless environment. In this dissertation, we propose two methods for improving performance in lossless compression. In comparison to HEVC, the proposed intra prediction and entropy coding achieve an improvement in compression performance as well as a reduction in complexity. The experimental results showed that the proposed intra prediction achieves a bit saving of 7.29% and a compression ratio 2.73 on average in the test sequences. And the proposed entropy coding achieves a bit saving of 1.46% and a compression ratio of 2.55 on average in the test sequences. When the proposed intra prediction and entropy coding are used together, a significant compression performance was achieved, with a bit saving of 7.99% and a compression ratio of 2.76. This performance was achieved through more accurate prediction and efficient coding of the coefficient by the proposed methods. In particular, the proposed intra prediction and entropy coding achieve significant compression performance in high-resolution sequences such as Ultra High Definition (UHD) and screen content, respectively.