Microarray image processing by order morphology

An order morphology transform is presented to filter and segment which is done by different percentile. Filter Is done flexibly by different size structure element with several percent. The threshold which for normal segment way such as Ostu decides is more lower when a low SNR Image Is processing especially the foreground is small or dot. The foreground can not be identified effectively in those case. Adaptive multl-threshold segment Is defined by percent value of order morphology. Analysis and results indicate that this way is more adaptive to different SNR fluorescence images. It could be applied to process high-density chips.

基于LabVIEW的VLC-LED模组结温非接触式测量系统设计

可见光通信中LED模组结温影响其光、色、电特性及工作寿命,对模组结温的测量及管理十分重要。因LED模组管脚存在无法接触测量的缺点,LED结温非接触式测量方法广受关注。在研究双光谱参数测量LED结温的基础上,开发了VC-LED模组结温的非接触式测量系统。该系统主要由恒流源控制、光谱数据采集和数据分析处理3个模块组成。经实验验证,双光谱参数法与中心波长法相比,两者结果偏差在3℃以内。采用该系统测量LED模组结温,不但直观、准确,且不需要接触LED管脚,实现了LED模组结温的非接触测量。

基于生成对抗网络与FACS的面部表情合成研究

为实现从单张图片合成连续变化的目标表情,提出一种基于生成对抗网络(GAN)和面部表情编码系统FACS(Facial Action Coding System)的表情合成方法。通过提取面部表情AU(Action Unit)作为生成表情的一种约束,利用生成对抗算法合成目标表情,这种结合表情编码的合成更具合理性。同时在网络中引入注意力机制,使网络在特征提取阶段更关注表情变化。实验表明,该方法能够克服图像中的光照和背景影响,合成表情具备连续性和真实性。该合成表情方法可应用于数字娱乐、影视制作等领域。