图像的多尺度表示指的是从原始图像出发,导出一系列越来越平滑、简化的图像。这种简化意味着信息的丢失。如果能定量描述每一个尺度中图像的信息,这对于多尺度表示来说有着重要的作用。虽然Sporring等人提出的尺度空间信息熵度量能解决一些问题,但是并不满足从视觉理论和直观的基础上提出的尺度空间信息量度量的基本要求,例如形态不变性等,为此在M arr视觉理论基础上定义了一个新的具有视觉意义的尺度空间信息度量,并在典型的高斯尺度空间中,证明了它确实满足从视觉理论和直观的基础上提出的尺度空间信息量度量的基本要求。数值试验验证了这种定义在视觉上是可靠的,从而为图像尺度的自适应选择提供了一种可靠的方法。

The basic idea behind a multi-scale representation is to embed the original image into such a one-parameter family of derived images,which become more and more smooth and simple.The simplification means the loss of details and information.It is important to measure the information of an image at a given scale and the loss between scales.Based on the vision theory and intuition of image processing,the paper proposed seven principles: nonnegativity,causality,Geometric invariability and so on.Sporring and Weickert have proposed a method of information measures.But Sporring's measure can not satisfy the principles proposed in this paper.This paper presented a new information measure based on the Marr's Vision Theory.In Gaussian Scale space of one dimension,we used the number of the first-class and the second-class of points as the information measure.Use the theory of Gaussian scale space,this paper has proofed that the new measure method satisfies the principles proposed.Then we extend the method in two dimensions directly.Experimental results proof its reliability.So this is a good choice of information meausures of scale-space based on the visual characters.