基于卷积神经网络的鸟类视频图像检索研究

本文介绍了在基于卷积神经网络的图像检索分析研究中,针对青海湖野鸟监控的视频关键帧图像数据,首先我们采用Vgg16神经网络的预训练模型来提取图像的特征,然后将特征向量作为图像的代表信息进行存储,构造图像的特征向量空间。之后对存储的高维特征向量进行降维处理,用主成分分析的方法将特征向量映射到低维的向量空间。最后应用余弦相似度算法对查询图像的特征向量与特征库中向量进行匹配,实现相似图像的检索。本文对提出的方法进行了实验,通过特征提取和恰当的特征降维,测试数据的检索准确率达到了89.82%。实验表明,本文提出的方法可以有效的实现鸟类图像的相似性检索。

The Research of Image Retrieval Based on Convolutional Neural Network

In the animation production process, In order to achieve real-time interaction between performers and role models, make the character animation's expression, mouth shape and sound consistent, and create a real natural role animation. In order to reduce the difficulty of facial animation, improve the quality and efficiency of animation production, a design scheme based on interactive character facial animation lipreading technology manufacturing method was proposed.