ID 原文 译文
38906 本文提出了一种改善深度修复图像统计特性一致性的方法。 A method of improving the statistical consistency for deep inpainted images is proposed.
38907 首先，分别采用非线性高通滤波残差及深度神经网络提取固有身份信号(intrinsic identity signal， IIS)，发现深度修复图像和真实图像存在IIS统计特性差异，验证在不同来源图像和不同的深度修复算法的条件下统计特性不一致性是普遍存在的。 Firstly, non-linear high-pass filtered residuals and a deep neural network are respectively used to extract intrinsic identity signal(IIS). It is found that there is IIS statistical inconsistency between the deep inpainted images and the pristine images, and such statistical inconsistency universally exists for different image sources and different deep inpainting algorithms.
38908 其次，提出一个生成型卷积神经网络，优化修复区域，保证修复图像的视觉质量，使其与真实区域IIS统计特性保持一致。 Secondly, a generative convolutional neural network is proposed to make the IIS statistics of inpainted regions be consistent with that of the pristine-regions, while maintaining high visual quality.
38909 最后，通过在合理范围内对生成网络的部分参数进行随机扰动，生成具有模式多样性的图像，有效降低生成图像被识别来源的概率。 Finally, by randomly perturbing some parameters of the generation network within a reasonable range, images with diverse patterns can be generated. It can reduce the detection accuracy of their sources being identified.
38910 通过对比真实图像、深度修复图像、生成图像的IIS统计特性，以及在取证检测器上的对抗检测实验，表明了本文方法的有效性。 The effectiveness of the proposed method is demonstrated by comparing the IIS statistics among pristine images, deep inpainted images and generated images, and evaluated by various kind of forensic detectors.
38911 为了解决图像密集字幕描述中感兴趣区域(Regions of interest，ROI)定位不准确与区域粗粒度描述问题，本文提出了一种基于深度卷积与全局特征的图像密集字幕描述算法， In order to solve the problems of inaccurate location of Regions of interest(ROI) and coarse-grained description of Regions in dense image cption, in this paper, an dense image description algorithm based on deep convolution and global features is proposed.
38912 该算法采用残差网络与并行LSTM(Long Short Term Memory)网络的联合模型对存在的区域重叠定位和粗粒度描述细节信息不完整问题进一步改进。 This algorithm adopts the joint model of Residual network and parallel LSTM(Long Short Term Memory) network to further improve the existing regional overlapping location and the incomplete coarse-grained description details.
38913 首先利用深度残差网络与Faster R-CNN(Faster R-Convolutional Neural Network)的RPN(Regional Proposal Network)层获取更精准区域边界框，以便避免区域标记重叠; Firstly, the depth Residual Network and the RPN(Regional Proposal Network) layer of Faster R-CNN are used to obtain more accurate regional boundary frame, so as to avoid overlapping of regional markers.
38914 然后将全局特征、局部特征和上下文特征信息分别输入并行LSTM网络且采用融合算子将三种不同输出整合以获得最终描述语句。 Then the global feature, local feature and context feature information are input into the parallel LSTM network respectively and the fusion operator is used to integrate the three different outputs to obtain the final description statement.
38915 通过在公开数据集上与两种主流算法对比表明本文模型具有一定优越性。 Compared with two mainstream algorithms on the open data set, the model presented in this paper has some advantages.