Changfei Zhou, Kai Lu, Wanqing Chi, Xiaoping Wang, Zhenhai Xiong National University of Defense Technology Changsha, China zcf337889342@163.com Abstract—Image processing has important influence during the development of visualization. With the scale and the complexity of image data expanding, it is becoming more and more difficult for image processing and target analysis. Saliency area detection of image processing, especially, visual attention is playing an important role the decreasing of computing burden and the improvement of the accuracy. Among this kind of algorithms, saliency detection based of structural similarity theory has more excellent performance than others through a new center-surround operator. In this paper, we realize the parallelization on data level and thread level of this algorithm on the MIC platform. The experiment results show that we could obtain a speed-up ratio of 5.6 in average with the help of the SIMD. We could also get a top performance speed-up ratio of 2.42 in average on an MIC coprocessor against a prevalent CPU.
[1]
Sunil D. Sherlekar.
Intel Many Integrated Core (MIC) Architecture.
,
2012,
ICPADS 2012.
[2]
A. Bovik,et al.
A universal image quality index
,
2002,
IEEE Signal Processing Letters.
[3]
Christof Koch,et al.
A Model of Saliency-Based Visual Attention for Rapid Scene Analysis
,
2009
.
[4]
Zhu Chuan-qi.
Partial Reuse of the Vector Registers in SIMD Optimization
,
2007
.
[5]
Eero P. Simoncelli,et al.
Image quality assessment: from error visibility to structural similarity
,
2004,
IEEE Transactions on Image Processing.