Layered video transmission based on genetic programming for lossy channels

Abstract

This paper presents a novel robust layered video transmission design algorithm for noisy channels. In the algorithm, the 3D SPIHT coding technique is used to encode the video sequences for the transmission of each layer. A new error protection allocation scheme based on genetic programming is then employed to determine the degree of protection for each layer so that the average distortion of the reconstructed images after transmission can be minimized. Simulation results show that, subject to the same amount of redundancy bits for error protection, the new algorithm outperforms other existing algorithms where equal-protection schemes are adopted.

Introduction

In many applications, digital video sequences are required to be transmitted over wireless or computer networks. Users in the networks might have different demands on the quality of the video sequences. Because of the variety of requests in the networks, it might be difficult for a server to provide an encoded bit stream satisfying all the requirements. One solution to this problem is to use the simulcast technique where video sequences are encoded and stored independently for each specific request. This approach requires more resources to be used in the encoder in terms of disk space and management overhead.

To eliminate this drawback, a number of layered transmission design algorithms [2], [7], [5] has been proposed. In the schemes, an encoded bit stream is delivered in more than one layer. By decoding bit streams accumulated up to different upper layers from the base layer, we can reconstruct the transmitted images with different qualities. Users with various requirements therefore share the same system for efficient video transmission. Many techniques such as embedded zero-tree wavelet (EZW) [10], 2D set partitioning in hierarchical trees (2D SPIHT) [9] and JPEG2000 [11] can be employed for realizing layered systems. Nevertheless, these techniques are used only for still images. They may not be well suited for the video sequences having high temporal correlation. The 3D SPIHT [6], which is an extention of the 2D SPIHT for video coding, can be used to solve this problem. The algorithm uses 3D wavelet transform for exploiting both the spatial and temporal correlations. It also provides the embedded bitstreams for attaining layered transmission.

Although the 3D SPIHT is effective for the implementation of layered transmission systems, its performance can be severely degraded when the bitstreams are delivered over lossy channels. A proper error protection scheme therefore is necessary for robust video transmission. The usual equal error protection schemes might not be suited for the SPIHT-based system. This is because the significance of bit streams delivered in different layers in the systems are not equal. The bit streams in lower layers are more important since an error occurs in these bit streams results in failure for decoding bit streams in higher layers.

In light of the facts discussed above, the objective of this paper is to present a novel unequal error protection scheme for the layered video transmission systems realized by the 3D SPIHT. This unequal protection problem is a typical combinatorial optimization problem which seeks the minimum of a given objective function of many variables. The variables interact with each other in a very complicated way, and their relationship usually cannot be expressed analytically.

One way to solve the problem is to use the simulated annealing (SA) [3], [13] technique. It mimics the objective function associated with the combinatorial optimization problem as the energy associated with the physical system. By slowly reducing an appropriately defined effective temperature, it seeks the minimum energy state as the solution for the optimization problem. Although the SA has been found to be effective, one has to be careful with the annealing schedule, the rate at which the temperature is lowered. The schedule achieving global optimum [3] requires tremendous computational complexities, and is not realistic in many applications. Other schedules that accelerate the cooling process can reduce the computational complexities at the expense of possible trapping in a poor local optimum.

The genetic algorithm (GA) [8] is another method for solving the combinatorial problem [1], [4]. It consists of a set of genetic strings where each string is a possible solution for the optimization problem. The regeneration, crossover, and mutation operations are then applied to these strings to obtain a better set of genetic strings. The same process is repeated until the algorithm converges to a nearby global optimum. The performance of GA is insensitive to the selection of algorithm parameters. In addition, since the algorithm is simple to implement, the computational complexities of GA is lower than that of SA. Because of these advantages, we use GA to realize the optimal error protection for layered transmission system. Simulation results show that, as compared with the techniques based on the simple equal protection scheme, the novel algorithm is less sensitive to delivery errors. The GA-based protection scheme therefore is an effective alternative for the applications where both superior rate distortion performance and robust transmission are desired.