A High Performance DCT Algorithm Based On The Stochastic Computation

With the widespread application of multimedia systems and real time communication systems in recent years, the effectiveness of transmission and storage for vast image data is becoming to the most critical factor in real time systems. Thousands of works have been done to solve this transmission and storage problem such as JPEG and MPEG. In JPEG and MPEG, discrete cosine transform (DCT) and inverse discrete cosine transform (IDCT) computation account for40percent of the composition and decomposition of JPEG and MPEG, which means that the effectiveness of DCT computation greatly impact on the computational complexity in JPEG and MPEG. Hence, we propose a novel DCT computing approach by combining the theorem stochastic computation and DCT butterfly algorithm to decrease the number of computation in traditional DCT.In this paper, we propose a high-performance DCT algorithm based on stochastic computation and the fast DCT algorithm. The main ideal of the research with this new approach can be divided into three units which will be mainly introduced by following parts.1) Image information to stochastic representation converter unit. The critical of the simplicity of stochastic computation is the representation of data in stochastic computation. In probability (stochastic) representation, the real-valued number x in unit interval is represented by a stochastic pulse stream X of length L which is a binary sequence and the weight of each bit in the random sequence are the same. In this random sequence X, each bit has the probability x of being1and probability1-x of being0. Therefore, the image information which represented in the format of binary radix should be mapped to the probability field firstly. According to a great number of experiments, the Gold pseudo-random number is suit for this the image information to probability representation conversion best.2) DCT computation with stochastic theory cell. According to the DCT definition and the existing calculation models of stochastic computation, we improve these calculation models to construct a new calculation model which is more appropriate for the fast DCT algorithm. Under the circumstances, the high-performance DCT algorithm based on stochastic computation is implemented excellently.3) Analysis of the improved method unit. During the implementing process, some elements make great influence to the result of the experiment. Here, the analysis of these factors is recommended. In order to verify the effectiveness of the algorithm, the experiments based on the Corel image database were carried out. Experimental result illustrates that the proposed approach on implementation of DCT which is simple-constructed has a lower energy cost and stronger error tolerance.