基于3D人脸跟踪的实时真人虚拟试发

真人虚拟试发在虚拟社交、在线商城试戴和个人发型设计等领域具有明显的应用价值.现有的基于图像的2D虚拟试发系统无法让用户从不同的角度和距离实时观看试发效果,极大地限制了这类系统的可应用范围.为此,提出一个基于3D人脸跟踪的实时真人虚拟试发系统,用户仅使用一个网络摄像头,就能如同在镜子前试戴真实假发一样进行虚拟试发.首先结合深度学习技术与传统优化算法,从2D人脸视频帧中快速鲁棒地计算出对应的3D人脸形状和姿态,实现3D人脸跟踪;然后根据事先指定的若干对对应点,自动把3D头发佩戴到视频帧所对应的3D人脸模型;最后根据3D人脸模型对3D虚拟头发进行遮挡处理与绘制,实现虚实混合的真人虚拟试发效果.此外,还提出一种简单实用的3D头发重上色算法,以进一步增强虚拟试发效果.在室内外环境下,使用网络摄像头采集用户视频进行了大量的真人虚拟试发实验的结果表明,该系统是有效的和实用的.

基于YUV颜色空间GAN网络的图像去雾算法研究

针对目前单幅图像去雾算法存在有色差,去雾效果不理想等问题,提出了一种基于YUV颜色空间的单幅图像去雾算法。该方法应用了GAN图像着色任务的思想,从正向的角度对雾霾图像实现重新上色。将雾霾图像转换至YUV颜色空间,在Y通道采用密集残差模块采集图片的亮度特征,根据特征对雾霾图像的亮度信息进行调整,降低雾霾对图像的影响。在UV通道上采用4个残差模块对图像颜色信息进行多次提取,根据提取的颜色信息通过模型预测对图像进行重上色。采用包含跳跃连接结构的特征融合网络将底层特征与高层特征进行融合,在融合过程中加入注意力模块以实现更加精细的去雾。实验结果表明,该算法在合成雾霾图像数据集上,RMSE,SSIM和PSNR 3种指标均达到了较高的水平,在真实雾霾图像上,对浓雾和薄雾图像均表现出了优异的去雾效果,具有良好的稳定性。

Impacts of Primo MAXX Growth Regulator on the Growth of Bermuda Grass

[Objective] The aim was to investigate the effects of Primo MAXX growth regulator on growth of Bermuda grass,explore appropriate application concentration of Primo MAXX on Bermuda grass,and provide basis for the application of Primo MAXX in Bermuda meadow.[Method]The height,leaf colour and the aerial part＇s fresh weight of Bermuda grass treated with different concentrations were researched using single factor random block test.[Result]Primo MAXX application effectively reduced the height in all treatment without significant difference;appropriate application of Primo MAXX can enrich the color;and 35 days after application,Bermuda grass roots were become thicker.[Conclusion]The application of Primo MAXX with effective dose of 150-300 g/hm2 on Bermuda grass in southern China is more appropriate.