基于跨层卷积神经网络的石刻碑文识别

Stone Inscription Recognition Based on Cross-Layer Convolutional Neural Network

石刻碑文的研究具有重要的历史价值和文化价值,但是由于受到自然环境的风化、腐蚀和人为的破坏,碑刻文字存在很大的识别难度.传统的识别方法通过选择特征的提取方式,得到笔画、部首等浅层特征,准确率不高.卷积神经网络可以利用深层网络自动提取到更抽象的特征用于识别,具有更高的准确率.本文针对碑刻数据收集困难,数据量较少的问题,提出了一种基于VGG-16的跨层卷积神经网络进行石刻碑文的识别,使用投影分割改进法实现碑文的分割,该方法在本文构建的测试数据集上对石刻碑文的识别具有很好的效果.

The study of stone inscriptions has important historical and cultural values,but due to the weathering,corrosion of the natural environment and man-made destruction,the inscriptions have difficulty to recognize.The traditional recognition method obtains shallow features such as strokes and radicals by the extraction method of electing the features,but the accuracy is not high.Convolutional neural networks can automatically extract more abstract features for identification by using deep networks,with higher accuracy.Based on the difficulty of collecting inscription data and the small amount of data,this paper proposes a cross-layer convolutional neural network based on VGG-16 to identify the stone inscription.The projection segmentation method is used to realize the segmentation of the inscription.This method is constructed in this paper.This method has a good effect on the identification of the stone inscriptions on the test data set constructed in this paper.