Scalable Video Coding Using Compressive Sensing

Abstract

the client has ample computational resources and a high resolution display, there is no way to get a better viewing experience which is commensurate with the available resources. This phenomenon is known as the "cliff effect"-no video is available unless some threshold constraints are met and no improvement is achieved when the constraints are exceeded. Scalable video coding (SVC) [18] encodes video into ordered layers, where each higher layer provides a refinement to the encoding of the lower layers. Since decoding of lower layers is a prerequisite for the decoding of higher ones, the lower layers need a higher level of protection in transmission. In broadcast applications, each layer may be transmitted as a separate stream, with different protection levels, e.g., by using hierarchical modulation [9]. In point-to-point applications, feedback from the client usually indicates how many layers are to be transmitted. In these situations, providing different protection to each layer is usually difficult-all layers receive the same protection, which severely limits the scalability. The practical limitation of SVC is its relatively small number of layers (due to

An Efficient Augmented Lagrangian Method with Applications to Total Variation Minimization

“…The theory indicates that a sparse signal under some basis may still be recovered even though the number of measurements is deemed insufficient by Shannon's criterion. Nowadays, CS has been widely studied and applied to various fields (see [22,10,21,13,38,39] for example). Given measurements b, instead of finding the sparsest solution to Ax = b by a combinatorial algorithm, which is generally NP-hard [25], one often chooses to minimize the 1 -norm or the total variation (abbreviated TV) of x.…”

Total Variation Applications in Computer Vision

“…TV methods also work well with piecewise constant images. Furthermore, since images are easier to compress when the discrete gradient representation is used, the TV-norm has advantages over wavelets in the presence of additive and/or quantization noise (Jiang, Li, Haimi-Cohen, Wilford, & Zhang, 2012).…”

Compressive measurements generated by structurally random matrices: Asymptotic normality and quantization

“…video surveillance and streaming [10][11][12][13][14][15][16] involve sending the measurements for processing over a communication channel. The transmission of measurements requires a coding scheme, which entails source coding that is typically implemented by quantization followed by channel coding of the quantization codewords.…”