基于改进深度学习的海量多媒体图像信息快速检索研究  

Fast Retrieval of Massive Multimedia Image Information Based on Improved Deep Learning

作　　者：高小芳 GAO Xiao-fang(Quanzhou Institute of Information Engineering,Quanzhou,Fujian 362000,China)

机构地区：[1]泉州信息工程学院,福建泉州362000

出　　处：《河北北方学院学报（自然科学版）》2022年第3期20-25,共6页Journal of Hebei North University:Natural Science Edition

基　　金：福建省教育厅中青年教师教育科研资助项目:“基于“SPOC+项目竞赛”模式的在线教学平台”(JAT200800)。

摘　　要：目的为了更为快速、准确地完成图像检索,提出一种基于改进深度学习的海量多媒体图像信息快速检索方法。方法首先对多媒体图像进行预处理,包括混合滤波去噪和直方图均衡化,以此提升图像质量,然后对深度学习中的卷积神经网络进行改进,增加图像特征融合层,避免图像细节特征丢失,最后通过度量图像特征之间的相似度来完成图像信息检索。结果实验结果表明,与传统方法相比,上述方法的图像检索平均精度均值更高、平均运行时间更短。结论证明所研究方法能在更短的时间内完成更为准确的检索,检索性能较好。Objective To complete image retrieval more quickly and accurately,a fast retrieval method of massive multimedia image information based on improved deep learning was proposed.Methods Firstly,the multimedia image was preprocessed,including hybrid filtering denoising and histogram equalization,so as to improve the image quality.Secondly,the convolution neural network in deep learning was improved,and the image feature fusion layer was added to avoid the loss of image detail features.Finally,the image information retrieval was completed by measuring the similarity between image features.Results The experimental results showed that compared with the traditional methods,the average precision of the research method was higher and the average running time was shorter.Conclusion It is proved that the method can complete more accurate retrieval in a shorter time with better retrieval performance.

关 键 词：深度学习 改进卷积神经网络 多媒体图像 特征提取 检索方法 

分 类 号：TP241.1[自动化与计算机技术—检测技术与自动化装置]