基于卷积神经网络的高相似度图像识别方法

A High Similarity Image Recognition Method Based on Convolutional Neural Network

考虑到部分图像中的相似度较高,会造成错误的图像分类或图像混淆。为了满足对高相似度图像的识别需求,提出一种基于卷积神经网络的高相似度图像识别方法。根据高相似度图像的预处理,计算图像特征的相似度加权和,通过构建高相似度图像特征的频数直方图,检索高相似度图像特征的相似度。基于神经元中图像特征之间的灵敏度,建立误差函数,结合采样层中图像特征的灵敏度,更新卷积神经网络的权值。利用迭代分析法,确定图像的中心点,以中心点为判定条件,实现高相似度图像的识别。实验结果表明,文中方法能够识别出具有较高相似度的图像,并提高图像的识别效率。

Considering the high similarity in some images,it can lead to incorrect image classification or image confusion.In order to meet the requirements of high similarity image recognition,a high similarity image recognition method based on convolutional neural network is proposed.Based on the preprocessing of high similarity images,it will calculate the weighted sum of similarity of image features,and retrieve the similarity of high similarity image features by constructing a frequency histogram of high similarity image features.Based on the sensitivity among image features in neurons,the error function is established,and combined with the sensitivity of image features in the sampling layer,the weights of convolutional neural network are updated.Using iterative analysis method,it will determine the center point of the image,and use the center point as the judgment condition to achieve high similarity image recognition.The experimental results show that the proposed method can recognize images with high similarity and improve the recognition efficiency of images.