SLIC算法的线程级并行优化研究与实现  

Research and Implementation of Parallel Optimization of SLIC Algorithm

作　　者：尚小敏 李强 齐永孟 陶顺安 SHANG Xiao-min;LI Qiang;QI Yong-meng;TAO Shun-an(College of Computer Science and Technology,Qingdao University,Qingdao 266071,China)

机构地区：[1]青岛大学计算机科学技术学院,青岛266071

出　　处：《青岛大学学报（自然科学版）》2022年第4期20-25,32,共7页Journal of Qingdao University(Natural Science Edition)

基　　金：山东省自然科学基金面上项目(批准号:ZR201910310143)资助。

摘　　要：为解决SLIC算法计算量大和数据依赖性强等问题,调整算法结构,将核心计算部分的基准点由聚类中心转为普通像素点,以计算量换取并行度,消除了像素点间的依赖;针对RGB转换Lab部分以及像素间的梯度计算部分,设计了基于OpenMP的多线程并行优化方法。通过整合并行方式实现了SLIC超像素并行分割算法,实测最终加速比达到29.0,相比对传统SLIC算法具有较为明显的加速效果。The algorithm structure adjustment was to solve the calculation burden and data dependency of the SLIC algorithm.The reference points of the core calculation part were transferred from the clustering center to ordinary pixel points.The dependence between pixels was eliminated by exchanging the amount of computation for parallelism.A multithreading parallel optimization method based on OpenMP was designed for the RGB conversion experiment and gradient calculation between pixels.The SLIC superpixel parallel segmentation algorithm is implemented by integrated parallelization,and the final speedup ratio is 29.0,which has a more obvious speedup effect compared with the traditional SLIC algorithm.

关 键 词：数字图像处理 超像素 并行计算 

分 类 号：TP338.6[自动化与计算机技术—计算机系统结构]