Cubic Video Coding using 3-D Wavelet Transform

在本研究中，我们提出了一种新的基于相位运动估计的三维视频编码方案，以及新的三维EZW-IP编码器。在MPEGIH.26x等传统的视频编码方法中，采用了运动估计和DCT变换的混合编码方案。但实际上，这些方法只使用几个帧的时间相关性。我们在这里注意到，并旨在使用更多的时间相关性来减少冗余。采用基于相位的运动补偿方法对连续帧之间的运动进行有效补偿。此外，该方法不需要为每个输入图像逐帧制备参考图像。收集8个预测错误帧，并生成一个CUBE。采用三维EZW-IP方法进行编码。我们的编码方案提供了一些理想的特性，包括精确的比特率控制、质量和分辨率可扩展性、渐进传输和良好的压缩性能。此外，我们提出了一种新的基于相位相关的四叉树结构运动估计方法，以减少预测误差。在传统的视频编码方案中，采用了块匹配运动补偿和频率变换。然而，传统的BM方法在运动补偿部分存在计算量大的问题。为了减少计算时间，提出了一种新的四叉树结构相相关运动估计方法。相位相关法比BM法计算效率高。四叉树结构不仅减少了运动补偿误差，而且降低了计算复杂度。通过一些实例，与BM方法相比，我们提出的方法具有良好的压缩性能和减少计算时间。

In this research, we present a new three dimensional video coding scheme based on a Phase-based motion estimation, and new three dimensional EZW-IP coder. In the conventional video coding methods like MPEGIH.26x, hybrid coding scheme using motion estimation and DCT transform is adapted. However actually, these methods use only the temporal correlation of several frames. We pay attention here, and aimed to use much more temporal correlation to reduce the redundancy. The motions between successive frames are compensated effectively by phase-based motion compensation method. Moreover, it is not necessary to prepare the reference image for each input image frame by frame in this proposed method. 8 predicted-error frames are collected, and one CUBE is generated. And it is encoded by 3-D EZW-IP method. Our coding scheme provides some desirable features including exact bit-rate control, quality and resolution scalability, progressive transmission, and good compression performance.Furthermore, we propose a new quadtree-structured motion estimation method based on a phase correlation to reduce the prediction error. In the conventional video coding scheme, they use block-matching(BM) motion compensation and the frequency transform. However the problem of the conventional BM method in the part of motion compensation is the computational demanding task. We aim to reduce the computational times and present a novel quadtree-structured phase correlation motion estimation. Phase correlation method is very computationally efficient than the BM method. Quadtree hierarchy scheme reduces not only motion compensation errors but also the computational complexity. Through some examples, our proposed method provides a good compression performance and reduction of computational times as compared to BM method.