Recognition Using a Multi-Channel Vision System and Its Application

我们开发了一种多通道视觉系统技术及其在视觉问题中的应用。多通道视觉系统在400-700 nm的可见光范围内具有三个以上的通道，不同于具有三个RGB通道的固定系统。这个系统对解决几个视觉问题很有用。首先，它提高了相机系统的颜色分辨率。其次，该系统使估计光谱函数变得容易，例如光源的光谱能量分布和物体表面的光谱反射率。多通道成像的基本系统由窄带滤光片、单色数码相机和个人计算机组成，在本研究中，我们首先开发了多通道成像系统和光谱信息估计算法。接下来，我们开发了一个预测物体三维反射特性的系统。此外，将多通道视觉系统应用于以下实际问题：(1)研究了一种识别自然场景中物体的有效方法和一种对原始电路板上的物体材料进行分类的方法。(2)提出了一种利用窄带滤波视觉系统对荧光场景光源进行分类的方法。大多数荧光光源光谱可以分为三类。(3)提出了一种基于库贝尔卡-蒙克理论的粉底肤色建模和表面光谱反射率估计方法。(4)提出了一种利用计算机视觉和图形学的新技术来实现艺术绘画数字化档案的方法。将数字存档过程分解为图像采集和图像渲染两个阶段。(5)提出了一种从摄像机瞄准镜面球拍摄的图像中估计场景光源光谱功率分布的全方位分布的方法。

We have developed a technique for multi-channel vision system and its application to vision problems. The multi-channel vision system has more than three channels in the visible range of 400-700 nm, different from the fixed system with three channels of RGB. This system is useful for solving several vision problems. First, it improves color resolution of the camera system. Second, the system makes it easy to estimate spectral functions such as spectral-energy distribution of a light source and spectral reflectance of an object surface. The basic system of multi-channel imaging consists of narrowband filters, a monochrome digital camera, and a personal computer.In this research, we first developed systems for multi-channel imaging and algorithms for estimating the spectral information. Next, we developed a system for predicting three-dimensional reflection properties of an object. Moreover, the multi-channel vision system was applied to the following practical problems.(1) We considered an effective method for identifying objects in a natural scene and a method for classifying object materials on a raw circuit board.(2) A method was developed for classifying fluorescent scene illuminant by using a vision system with narrow band filtration. Most fluorescent illuminant spectra can be classified into three groups.(3) A method was devised a method for modeling human skin coloring with foundation makeup and estimating the surface-spectral reflectance by using the Kubelka-Munk theory.(4) We proposed an approach to digital archives of art paintings by using new techniques of computer vision and graphics. The process of digital archiving is decomposed into two stages of image acquisition and image rendering.(5) A method was developed for estimating an omnidirectional distribution of the scene illuminant spectral-power distribution from images taken by a camera aimed at a mirrored ball.