翻译A Multiscale Retinex for Bridging the Gap Between Color Images and the Human Observation of Scenes

A Multiscale Retinex for Bridging the Gap Between Color Images and the Human Observation of Scenes

缩小彩色图像和人类视觉场景差距的一种多尺度Retinex方法

Abstract— Direct observation and recorded color images of the same scenes are often strikingly different because human visual perception computes the conscious representation with vivid color and detail in shadows, and with resistance to spectral shifts in the scene illuminant. A computation for color images that approaches fidelity to scene observation must combine dynamic range compression, color consistency—a computational analog for human vision color constancy—and color and lightness tonal rendition. In this paper, we extend a previously designed single-scale center/surround retinex to a multiscale version that achieves simultaneous dynamic range compression/color consistency/lightness rendition. This extension fails to produce good color rendition for a class of images that contain violations of the gray-world assumption implicit to the theoretical foundation of the retinex. Therefore, we define a method of color restoration that corrects for this deficiency at the cost of a modest dilution in color consistency. Extensive testing of the multiscale retinex with color restoration on several test scenes and over a hundred images did not reveal any pathological behavior.

摘要：由于人类视觉感知系统在接受色彩时有意识的透过阴影，恢复其本来的色彩，所以对于相同的场景，人们直接观察到的和计算机记录的彩色图像往往有较大的差别。彩色图像，接近忠实于现场的观察必须结合动态范围压缩，色彩一致性的计算模拟人类视觉的颜色恒常性和色彩和亮度色调再现。在本文中，我们延续先前设计的单级中心/环绕Retinex的到，实现同步动态范围压缩/色彩一致性/亮度再现一个多尺度的版本。这个延伸无法产生良好的色彩还原为一个类，它包含违反灰色世界假设隐到了Retinex的理论基础的图像。因此，我们定义色彩还原的方法来解决这方面的不足在温和的稀释色彩一致性的成本。多尺度Retinex的颜色恢复对几个测试场景和百余影像广泛的测试，并没有发现任何病态行为。