面向视频超分辨率重建的混合粒子群优化算法  

A Hybrid Particle Swarm Optimization Algorithm for Image/Video Super-resolution Reconstruction

作　　者：俞文静[1] 张明军[1] 王影[1] YU Wen-jing;ZHANG Ming-jun;WANG Ying(South China Institute of Software Engineering,Guangzhou 510990,China)

机构地区：[1]广州大学华软软件学院,广东广州510990

出　　处：《计算机技术与发展》2018年第11期75-78,84,共5页Computer Technology and Development

基　　金：2015年广东省高校省级重点平台和重大科研项目(2015KQNCX202);广州大学华软软件学院科研项目(ky201723)

摘　　要：针对多帧视频序列图像超分辨率重建问题中图像序列的像素相关性,建立了一种有效的视频图像超分辨率重建数学模型,将视频超分辨率重建问题转化为从低分辨率图像像素序列到高分辨率图像像素序列的寻优问题。针对基本粒子群(PSO)算法的缺陷,将遗传原理及蚁群机制引入PSO,提出了一种改进的MPSO算法。结合视频序列超分辨率重建问题特性,图像/视频可以表示为粒子像素序列,并且相邻视频帧之间序列具有规则性、相关性以及约束性等特性。重新定义了算法的粒子的输入、粒子交叉、粒子变异以及粒子的位置和速度运动方程,并设定了粒子优化目标以及运动适应度准则。最后通过典型的室内监控人物运动的视频图像超分辨率重建实例,验证了提出的模型和算法的可行性。Aiming at the pixel correlation of image sequences in super-resolution reconstruction of multi-frame video sequence,we establish an effective image/video super-resolution reconstruction mathematical model.The problem of multi-frame video’s super-resolution reconstruction is transformed into the optimization problem from low resolution image to high resolution image.Aiming at the defects of basic particle swarm optimization(BPSO),we introduce the genetic principle and ant colony mechanism into PSO and propose an improved PSO algorithm,called mixed PSO(MPSO).Combined with the characteristics of super resolution reconstruction of video sequences,images/videos can be represented as particle pixel sequences,and the sequences between adjacent video frames have regularity,correlation and constraints.MPSO,including its particle input,particle crossover,particle variation,particle position and motion equations,optimization objective and fitness criterion,are designed.Finally,the experiment of typical indoor surveillance video super-resolution reconstruction verifies the feasibility and effectiveness of the proposed model and algorithm.

关 键 词：图像 视频 超分辨率重建 粒子群算法 优化算法 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构] TP391.9[自动化与计算机技术—计算机科学与技术]