基于特征融合的商标检索方法

Trademark retrieval method based on feature fusion

各种各样的字体导致卷积神经网络对商标图像提取的特征存在冗余,文字部分的信息不仅没有充分利用,反而对图像特征的匹配产生了影响。提出一种融合卷积特征与文字特征的商标检索方法,在深度学习检索方法的基础上融合文字特征匹配的结果。针对文字特征之间的距离计算,提出一种改进的编辑距离算法用于提高文字特征的匹配效果,提高商标检索的准确率。

A variety of fonts leads to the redundancy of features extracted from trademark images by convolutional neural network.The text information is not fully utilized,but has an impact on the matching of image features.A trademark retrieval method combining convolution feature and literal feature was proposed.Based on the deep learning retrieval method,the results of literal feature matching were fused.An improved edit distance algorithm was proposed to improve the matching effect of text features,so as to improve the accuracy of trademark retrieval.