Parallelism Research For Multimedia Retrieval Algorithm

With the rapid popularization of computer and Internet, multi-media data has become one of the major data types processed in the Internet. But the multimedia data contains too much information, the current multimedia retrieval algorithm cannot retrieval this data efficiently, makes many multimedia applications are far from applicable. There are mainly two classes of multimedia retrieval algorithms:one is based on global features and the other is based on local features. The global feature algorithms use a unique feature to represent an image or one frame in a video. Though they process the image fast, but the low precision rate makes they are not suitable for current multimedia applications. The local feature algorithms use hundreds of features to describe an image, which leads to high precision rate for multimedia retrieval algorithm. However, the large quantity of data and calculation of the local feature algorithms leads they cannot satisfy the real-time processing requirements. Therefore, the acceleration of multimedia retrieval algorithms based on local features becomes the focus.Currently, a variety of parallel hardware architecture (multi-core processors, GPUs and so on) has become more and more popular. Therefore, parallel characteristics analysis for multimedia retrieval algorithm can not only accelerate the current multimedia applications, but also give some references to future architecture research and design for these applications. The contributions of this paper are:●We analyze and implement the parallelism scheme on the local feature multimedia retrieval algorithm, and the experiments show that:■Block parallelism scheme can get the good speed up on most parallelization, and pipeline get the best scalability effect.■On the commodity 16-core machine, the best parallel implementation achieves a speedup of 13.5X.●We analyze the result of the experiments and summarize several design points when parallelism the local feature algorithms, provides some useful advice for future architecture design for multimedia applications.