深度卷积特征在素描作品分类与评价中的应用
发布时间：2019-05-08 03:43
【摘要】：以素描教学过程中的临摹作品作为研究对象,将深度卷积特征应用于素描作品的分类与评价中.首先测试深度卷积特征在素描作品分类中的效果,同时将素描作品评价问题转换为基于作品的构图、形准、质感、画面整体黑白灰等图像高阶语义特征的细分类问题(优、良、中、差);然后提出双线性卷积模型,以较好地解决图像细分类问题;最后使用Tensor Sketch投影算法将双线性深度卷积特征进行压缩,并采用端到端的训练进行模型微调.实验结果表明,在素描作品分类任务中,深度卷积特征明显优于传统手工特征(如直方图特征、纹理特征和SIFT特征);在素描作品评价中,压缩的双线性深度卷积特征能在较低维度上达到相似的评价效果.
[Abstract]:The deep convolution feature is applied to the classification and evaluation of sketch works by taking the copying works in the process of sketch teaching as the research object. In this paper, we first test the effect of deep convolution feature in the classification of sketch works. At the same time, the evaluation problem of sketch works is transformed into the fine classification of high-order semantic features of images such as composition, accuracy, texture, whole picture, black and white, and so on. Medium, poor); Then the bilinear convolution model is proposed to solve the problem of image fine classification. Finally the bilinear depth convolution feature is compressed by Tensor Sketch projection algorithm and the end-to-end training is used to fine-tune the model. The experimental results show that the depth convolution feature is superior to the traditional manual feature (such as histogram feature, texture feature and SIFT feature) in the classification of sketch works. In the evaluation of sketch works, the compressed bilinear deep convolution features can achieve similar evaluation results in lower dimensions.
【作者单位】： 浙江大学计算机科学与技术学院;
【基金】：国家自然科学基金(61562072,61303137,61402141) 教育部博士点基金(20130101110148)
【分类号】：TP391.41
本文编号：2471592
[Abstract]:The deep convolution feature is applied to the classification and evaluation of sketch works by taking the copying works in the process of sketch teaching as the research object. In this paper, we first test the effect of deep convolution feature in the classification of sketch works. At the same time, the evaluation problem of sketch works is transformed into the fine classification of high-order semantic features of images such as composition, accuracy, texture, whole picture, black and white, and so on. Medium, poor); Then the bilinear convolution model is proposed to solve the problem of image fine classification. Finally the bilinear depth convolution feature is compressed by Tensor Sketch projection algorithm and the end-to-end training is used to fine-tune the model. The experimental results show that the depth convolution feature is superior to the traditional manual feature (such as histogram feature, texture feature and SIFT feature) in the classification of sketch works. In the evaluation of sketch works, the compressed bilinear deep convolution features can achieve similar evaluation results in lower dimensions.
【作者单位】： 浙江大学计算机科学与技术学院;
【基金】：国家自然科学基金(61562072,61303137,61402141) 教育部博士点基金(20130101110148)
【分类号】：TP391.41
【相似文献】
相关期刊论文 前1条
1 莎莎;;素描作品“拍”出来[J];电脑爱好者;2014年17期
，本文编号：2471592
本文链接：https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/2471592.html
