Postdoctoral Research Associateship: Computer Visual Analysis of Surface Reflection

表面反射特性的视觉测量是计算机视觉基础研究和应用研究中的一个重要问题。表面反射率为图像分割和目标识别提供了可靠的线索。提出的研究目标是基于物理模型从朗伯反射中检测和分离镜面反射率。为了提供足够的信息来提取反射率的各种属性，我们通常需要一帧以上的图像。一个方向是使用颜色信息，另一个方向是使用具有不同视图的多个图像。提出了一种彩色图像分割及镜面反射和互反射检测算法。除了使用颜色，还提出了使用不同的视图来区分镜面反射率和拉梅尔反射率，并获得对象结构。该算法基于这样一个事实，即镜面反射度的外观随观察方向而变化，而Lambertian的镜面反射度外观则不随观察方向变化。为了建立一个更健壮和可靠的视觉系统，将按颜色和多个视图整合信息。为了有效地处理大量的光谱(颜色)、时间(多视图)和空间(图像)数据，将设计并实现使用4K处理器的连接机的并行算法。高度的计算并行性将使算法成为现实。

Visual measurement of surface reflectance properties is an important issue both in basic and in applied computer vision research. Surface reflectance gives a reliable cue for image segmentation and object recognition. The goal of the proposed research is to detect and separate specularity from Lambertian reflectance based on physical models. In order to provide enough information to extract various properties of reflectance, we generally need more than one frame of images. One direction is to use color information and another is to use multiple images with different views. An algorithm for color image segmentation and detection of specularity and inter-reflection is proposed. In addition to the use of color, the use of different views to differentiate specularity from Lamertian reflectance and to obtain object structure is proposed. the proposed algorithm is based on the fact that the appearance of specularity varies depending on viewing direction while that of Lambertian does not. In order to build a more robust and reliable vision system, information by color and multiple views will be integrated. For efficient processing of a large amount of spectral (color), temporal (multiple views) and spatial (image) data, parallel algorithms using a Connection Machine of 4K processors will be designed and implemented. The high degree of computational parallelism will bring the algorithms into practical reality.