基于视频自适应采样的快速图像检索算法  

A Fast Image Retrieval Algorithm Based on Video Adaptive Sampling

作　　者：谭文斌 黄贻望 刘声[1] TAN Wenbin;HUANG Yiwang;LIU Sheng(College of Data Science,Tongren University,Tongren 554300,China;Guizhou Provincial Key Laboratory of Public Big Data,Guizhou University,Guiyang 550025,China)

机构地区：[1]铜仁学院大数据学院,贵州铜仁554300 [2]贵州大学贵州省公共大数据重点实验室,贵州贵阳550025

出　　处：《软件导刊》2023年第7期131-137,共7页Software Guide

基　　金：国家自然科学基金项目(62066040,62261047);贵州省教育厅项目(黔教合KY字〔2022〕070号);铜仁市科研项目(〔2019〕97号,〔2022〕59号)。

摘　　要：为解决智慧农业监控系统目标图像检索计算量较大、耗时较长的问题,提出一种视频自适应采样算法。首先,根据视频相邻帧相似度变化情况自适应调整视频帧的采样率以提取视频关键帧,确保提取的关键帧能替代相邻帧参与目标图像检索计算。然后,将视频关键帧以时间为轴构建视频帧检索算子,代替原视频参与目标图像检索计算,从而减少在视频中检索目标图像时的大量重复计算,达到提升检索效率的目的。实验表明,自适应采样算法相较于固定频率采样、极小值关键帧算法所构建的视频帧检索算子检出率更高、更稳定。在确保图像被全部检出的基础上,使用视频帧检索算子替代原视频参与目标图像检索计算的优化幅度较大,时耗减少了60%以上,对提升智慧农业监控系统中目标图像的检索效率具有重要意义。To solve the problem of high computational complexity and time-consuming target image retrieval in smart agricultural monitoring systems,a video adaptive sampling algorithm is proposed.Firstly,adaptively adjust the sampling rate of video frames based on changes in sim⁃ilarity between adjacent frames to extract video keyframes,ensuring that the keyframes extracted by the algorithm can replace adjacent frames in target image retrieval calculations.Then,a video frame retrieval operator is constructed based on the time axis of the video keyframes,re⁃placing the original video to participate in the target image retrieval calculation,thereby reducing a large number of repeated calculations when retrieving the target image in the video,and achieving the goal of improving retrieval efficiency.Experiments have shown that the adaptive sam⁃pling algorithm has a higher and more stable detection rate than the video frame retrieval operator constructed by fixed frequency sampling and minimum keyframe algorithms.On the basis of ensuring that all images are detected,using video frame retrieval operators to replace the origi⁃nal video in the calculation of target image retrieval has a significant optimization range,reducing time consumption by more than 60%,and is of great significance for improving the retrieval efficiency of target images in smart agricultural monitoring systems.

关 键 词：自适应采样 图像相似度 目标图像帧 视频帧检索算子 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]