Game Development Reference
In-Depth Information
nonoverlapped patch. These coefficients, dictionaries, and DC values are quantized
and entropy coded to form the final bit stream.
Because of its promising performance in visual signal restoration, sparse repre-
sentationwas also applied as a postprocessingmodule in video compression. InXiong
et al. (
2013
), the authors proposed a sparse spatial-temporal representation scheme
for lowbitrate video coding. In this scheme, the key frames are compressed at original
resolution, and the rest are downsampled and reconstructed at the decoder. At the
decoder, using key frames as the training set, the nonkey frames can be restructured
by sparse-based superresolution.
12.3 Internet Media-Oriented Compression
With the development of Internet technology, cloud services have reached a maturity
that leads it into a productive phase. State-of-the-art video compression schemes
usually take advantages of the spatial, temporal, and statistical redundancy within
the video itself. The external similar video in the everywhere cloud provide more
opportunities in improving the coding efficiency by exploiting this external redun-
dancy. Specifically, the substantial increase of near-duplicate image and video sets
in Internet also provides us more redundancy reduction possibilities in image and
video compression. Yet, most algorithms in the literature are focusing on image
compression. In this chapter, we will review the Internet media-oriented compres-
sion techniques from three aspects: cloud-, set- and object-oriented image and video
compression.
12.3.1 Cloud-Based Compression
The main objective in visual signal compression is to compactly and faithfully repre-
sent the input signal at the receiver side. In addition to the pixel level representation
and reconstruction, local feature-based representation using large database has been
recognized to be a powerful method. The SIFT (scale-invariant feature transform)
descriptor proposed in Lowe (
2004
) present distinctive invariant features of images
by describing the local feature by location, scale, orientation, and histogram of gra-
dient directions. SIFT descriptors have been successfully been applied in object
recognition, content-based image retrieval, and other applications. In addition, other
feature descriptors have also been proposed for task specific applications. For exam-
ple, in Ji et al. (
2012
) the compact topical descriptor is proposed to learn the image
signature from the reference image corpus. In Bay et al. (
2006
), the SURF (Speed Up
Robust Feature) descriptor is proposed, which significantly reduces the dimensions
of feature vector and computational cost in describing an image.
With the development of mobile technology, such as high-resolution cameras,
powerful CPUs as well as 3G and 4G wireless communications, there is a high
