Research On Key Algorithm Of Ancient Figure Painting Image Segmentation Based On Deep Learning

In recent years,with the development of deep learning,image segmentation technology has made tremendous progress.There are a large number of cultural relic painting images in our country.Therefore,it is of great significance to study high-performance and high-accuracy image semantic segmentation algorithms in many fields such as cultural and creative industries and art appreciation analysis.However,due to the particularity of image data of cultural relics painting,the common structure of image segmentation neural network can not produce good results.In view of the above problems,this paper researches on the image segmentation of the ancient figure painting based on deep learning.The main algorithms and achievements of this paper are as follows:(1)A full-convolution neural network model based on dilated convolution and hypercolumns is proposed.Based on the full convolutional neural network structure,this paper introduces dilated convolution,and uses hypercolumns to integrate shallow-level features and deep-level features in convolutional neural networks.In this paper,a pixel classification module based on the large kernel convolution is designed to deal with the hypercolumns and obtain good segmentation results.(2)A fine grained image segmentation algorithm based on neural network and Graph-Cut algorithm is proposed.The result obtained by the neural network is relatively rough for the lines and edges in the ancient painting images.In view of the above problems,this paper improves the Graph-Cut algorithm by using convolutional neural network.The improved algorithm can correct the segmentation result of the image segmentation neural network and get the refined segmentation results of the ancient figure painting images.The proposed algorithm performs better than the neural network model such as SegNet according to the comparing experiments on Camvid dataset and PASCAL VOC 2012 dataset.The algorithm also achieves very good results on the ancient figures image dataset.It proves the effectiveness of the segmentation algorithm for the ancient figures images.