基于卷积神经网络和注意力机制的图像检索

Image Retrieval Based on Convolutional Neural Network and Attention Mechanism

当前先进的图像检索方法中,存在着不能很好地分辨图像中不同区域和内容的重要性的问题,导致计算资源分配不合理、检索正确率较低等一系列结果。为了解决这些问题,提出了一种基于卷积神经网络(Convolutional Neural Network,CNN)和注意力机制的图像检索方法。首先使用卷积神经网络提取特征,然后使用注意力机制处理提取的特征,可以在计算能力有限的情况下根据图像中的内容合理分配计算资源,使图像中的突出部分得到更多的关注。最后通过融合全局平均池化层处理后的CNN特征来进行图像检索。所提方法在corel1k、corel5k和corel10k三个数据集上与其他先进的图像检索方法进行了比较,结果表明该方法能够有效提高图像检索的精确率和召回率,并且在检索返回的前k张图像的数量(top-k)增加时,仍能保持良好且稳定的检索精确率。

The importance of different regions and contents in the image cannot be well distinguished in the current advanced image retrieval methods,which leads to the results like unreasonable allocation of computational resources and low accuracy.To solve these problems,this paper proposes an image retrieval method based on convolutional neural network(CNN)and attention mechanism.First,the CNN is used to extract features.Then the attention mechanism is used to process the extracted features,which can reasonably allocate computing resources according to the contents of the image in the case of limited computing power,so that the highlights in the image can get more attention.Finally,the CNN features processed by the global average pooling layer are fused to image retrieval.This method is compared with other advanced image retrieval methods on corel1k,corel5k and corel10k datasets.The experimental results show that this method can effectively improve the precision rate and recall rate of image retrieval,and can maintain a good and stable precision rate when the number of the first k images(top-k)increases.