• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文期刊>International Journal of Innovative Computing Information and Control >DEVELOPMENT OF A SKIN TEXTURE EVALUATION SYSTEM USING A CONVOLUTIONAL NEURAL NETWORK
【24h】

DEVELOPMENT OF A SKIN TEXTURE EVALUATION SYSTEM USING A CONVOLUTIONAL NEURAL NETWORK

机译：利用卷积神经网络开发皮肤纹理评估系统

获取原文
获取原文并翻译 | 示例
           
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

Currently, various methods are used to evaluate the beauty of skin, but problems arise with evaluation accuracy and the time required for evaluation. Previous research has proposed a method in which researchers select features and evaluate skin texture using machine learning with a three-layer neural network. However, results of this research were discrepant from visual evaluation by experts. In this paper, we describe the development of a convolutional neural network that acquires skin features through learning. We used images captured with a microscope as input data and visual evaluation scores as training data. To evaluate performance, we examined the correlation coefficient between the estimated evaluation score by the convolutional neural network and the visual evaluation score, as well as the matching rate of all data scores. When data from a single subject were used for learning but were not used as test data (an "open set"), the correlation coefficient and matching rate were 0.903 and 73.2%, respectively. Alternately, when data from a single subject were used for both learning and test data (a "closed set"), the correlation coefficient and matching rate were 0.922 and 77.8%, respectively. In each case, these values were significantly higher than those of previous studies.
机译：目前，各种方法用于评估皮肤的美丽，但出现问题，评价准确性和评估所需的时间。以前的研究提出了一种方法，其中研究人员使用三层神经网络使用机器学习选择特征和评估皮肤纹理。然而，这项研究的结果是由专家视觉评估的差异。在本文中，我们描述了通过学习获取皮肤特征的卷积神经网络的发展。我们使用用显微镜捕获的图像作为输入数据和视觉评估分数作为训练数据。为了评估性能，我们检查了卷积神经网络的估计评估分数与视觉评估评分的相关系数，以及所有数据分数的匹配率。当来自单个主题的数据用于学习但未用作测试数据（“开放式”）时，相关系数和匹配率分别为0.903和73.2％。或者，当来自单个主题的数据用于学习和测试数据（“闭合”）时，相关系数和匹配率分别为0.922和77.8％。在每种情况下，这些值明显高于先前研究的值。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱：kefu@zhangqiaokeyan.com

京公网安备：11010802029741号 ICP备案号：京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有

  • 客服微信

  • 服务号