Game Development Reference
In-Depth Information
from the same original video by different frame rate, resolution and quality. In each
video itself, there exist strong intercorrelations.
However, it is commonly known that better quality prediction will result in high-
efficiency video compression. Therefore, other near-duplicate video (NDV) still can
provide better prediction quality in video set compression. Inspired by multiview
video coding, in Wang et al. (
2014
) the authors proposed to compress the NDV with
a new coding framework, which includes four parts: analyzer, encoder, assembler and
decoder. In the analyzer, the NDVs and similar frames in the NDVs are identified. The
analysis information is passed to the encoder to construct the prediction structure. In
the encoder, similar to traditional MVC encoder, the current frame can choose the
reference frame from the video itself or the independent video, and rate distortion
optimization is used in this selection. The assembler generates each video into a
single bit stream and transmitted to the decoder.
Overall, the rapid developing cloud and Internet techniques have provided many
new opportunities for image and video coding, as more and more images and videos
will be stored in cloud in the future. In the cloud-based image and video processing,
pixel level comparison would not be as efficient as feature-based matching, as the
pixel-wise comparison could not be able to deal with the complex and various rela-
tionships in cloud, such as geometric deformation, shifting, illumination changing,
and spatial/temporal resolution scaling. Therefore, feature-based prediction scheme
will play more important roles in Internet video coding, and how to efficiently choose
and utilize the various features for compression will be extensively studied in the
future.
12.3.3 Visual Object-Based Coding
With the increasing development of Internet techniques and the growing huge amount
of data, there is a desired demand for the compact representation and high efficiency
compression of visual object in image and video content. The natural scene can be
decomposed into dynamic regions or visual objects with different spatial-temporal
characteristics, such as shape, color, and motion information. Different from current
block-based compression techniques, the visual object-based coding breaks the lim-
itations of the block size and boundary to achieve a more efficient representation and
higher coding performance. Moreover, the visual object representation in the cod-
ing techniques will dramatically accelerate the development of artificial intelligence,
because a wide range of applications rely on the object extraction and representation
such as object tracking and analysis.
The previous MPEG-4 was such an object-based standard that first standardized
the visual object representation in the coding stream. Themost recent HEVC standard
leads to a milestone that doubling the efficiency of the AVC/H.264. It is because
some key techniques have been involved in the novel HEVC standard. The variable-
size block is one of the most significant tool that allows the size of coding unit
