Sentiment Classification of Chinese Paintings via Feature Recalibration of Deep Network Aggregation

Analyzing and visualizing the sentiment of Chinese painting can effectively assist users in appreciating and learning.This paper proposes a Chinese painting sentiment classification algorithm based on feature recalibration of deep network aggregation.i)According to the unique characteristics of Chinese paintings,CNN is optimized to obtain better emotional features.Firstly,based on the seam carving technique,Chinese paintings are resized to preserve the brushwork while avoiding distortion.Secondly,the feature recalibration network module of multi-layered aggregation is proposed.The features in the block are aggregated and recalibrated.ii)The multi-class weighted activation mapping technique is proposed.The activations of each class are calculated respectively,and the weighted activation fusion is used to highlight the sentiment discriminative regions,thereby realizing the visualization of Chinese painting sentiment.The classification accuracy rate of 1000 Chinese paintings is 85.8%.Experimental results support that the proposed algorithm outperforms the existing representative benchmarks and locates the emotional regions correctly.