作者:卢克斌，殷守林
Authors:LU Ke-bin,YIN Shou-lin摘要:摘要:情感识别是计算机视觉研究中的一个热点，研究中国画表现的情感对于作品鉴赏工作具有重要意义。为了提高识别性能，针对传统卷积神经网络用于提取中国画的局部区域信息会导致有效信息丢失的问题，文章提出一种基于端到端弱监督学习网络方法对中国画情感进行识别。提出的学习网络由2个分类模块和1个情感强度预测模块组成。首先，在改进特征金字塔网络的基础上构建强度预测通道，提取多层次特征。使用基于梯度的类激活映射技术从第一个分类通道生成伪强度映射图，以指导提出的网络进行情感强度学习。将预测的强度图输入到第二分类通道中进行最终的中国画情感识别。最后，在公开数据集上对提出的方法进行了验证，实验结果表明，所提出的网络就混淆矩阵、平均分类准确率、平均情感识别率分别提高了10%，15%和13%。
Abstract:Abstract:Emotion recognition research is a hot spot in computer vision, and the study of Chinese painting emotion is of great significance to the appreciation of works. In order to improve the recognition performance, the traditional convolutional neural network used to extract local information of Chinese painting will lead to the loss of effective information. Therefore, the end-to-end weakly supervised learning network is proposed to recognize the Chinese painting emotion.The proposed learning network consists of two classification modules and one affective intensity prediction module.First, the intensity prediction flow is constructed on the basis of improved feature pyramid network to extract multi-level features.The gradient-based class activation map technique is used to generate pseudo-intensity maps from the first classification stream to guide the emotional intensity learning of the proposed network. The predicted intensity map is input into the second classification stream for the final Chinese painting emotion recognition. Finally, the proposed method is verified on the open data set.The experiment results show that the proposed network has improved the confounding matrix, average classification accuracy and average emotion recognition rate by 10%, 15% and 13% respectively.

PDF全文下载地址:

可免费Download/下载PDF全文