Design of A Parallel Algorithm for Visual Object Retrieval

• Main Authors: Yu-Jhih Liao, 廖于緻
• Other Authors: Shyi-Chyi Cheng
• Format: Others
• Language: zh-TW
• Published: 2006
• Online Access: http://ndltd.ncl.edu.tw/handle/55815118246736301830

碩士 === 國立高雄第一科技大學 === 電腦與通訊工程所 === 94 === Abstract In this thesis, a distributed and a parallel execution models for retrieving database images which contains the visual objects of a query image are presented. The problem to retrieve an image containing important visual objects from a large multimedia database efficiently remains as a challenge. Many methods proposed in the literature need modified to satisfy the real-time requirement of image retrieval. Many factors affect the retrieval accuracy and speed. For example, the visual object might be translated, rotated, and scaling from the database images and increase the difficulty for retrieval. The problems to retrieval visual objects from a large multimedia database are twofold: data skew and execution skew. The parallel architecture of the media server should be designed properly in order to overcome the I/O bound problem. Furthermore, the sequential execution model of general CPUs cannot efficiently explore the fine-gram parallelism of media computation in general. In this thesis, we design a parallel algorithm to speed up the execution of the proposed visual object retrieval method on the basis of the modified generalized Hough transform (GHT) and visual pattern matching. Experimental results show that the execution speed of the proposed III method is much better than that of the original visual pattern method proposed in our previous work.