Development and implementation of an improved diamond search algorithm based on adaptively thresholding

Meng Chun Lin*, Lan-Rong Dung

*Corresponding author for this work

Research output: Contribution to journalArticle

Abstract

In the present video compression standards, such as MPEG standards and H.26X, motion estimation occupies that most computational complexity in coding process, thus, many fast algorithms have been presented to save the computational load of motion estimation. Among these fast algorithms, the diamond search (DS) algorithm can efficiently supply better visual qualities under reducing the computational load for different video sequences, but it does not adaptively reduce unnecessary searching steps based on the content complexity of video sequence to significantly save computational power. Therefore, this paper presents an efficient early termination mechanism based on adaptively thresholding and this mechanism can efficiently avoid unnecessary searching steps according to the prediction result. The proposed thresholding not only has better adaptability for different video sequences but also can be easily combined with other fast motion estimation algorithms. Simulation results have been shown that four diamond search-like fast algorithms based on the proposed mechanism can efficiently save computational loads and maintain better visual-quality performance. Finally, we successfully implement a new diamond search architecture based on the proposed early termination mechanism and this architecture uses a special memory address generator to achieve low off-chip memory utilization and consumes 102 mW at 67 MHz in real-time processing.

Original languageEnglish
Pages (from-to)440-445
Number of pages6
JournalWSEAS Transactions on Circuits and Systems
Volume6
Issue number4
StatePublished - 1 Apr 2007

Keywords

  • Adaptively thresholding
  • Diamond search
  • Motion Estimation

Fingerprint Dive into the research topics of 'Development and implementation of an improved diamond search algorithm based on adaptively thresholding'. Together they form a unique fingerprint.

  • Cite this