论文部分内容阅读
在多光谱遥感数据中,用分类技术区分特征会产生带有“盐粒”和“胡椒粒”现象的图像,这种“嗓声”可以用逻辑平滑算子去掉,但会导致不希望的信息损失。用附加连通规则来约束逻辑平滑算子,使其对图像上的基本区域起作用,就可以获得较好的效果。本文描述了非约束和约束算子的性状,提出了一种算法,这种算法反复地应用约束算子,直到图像中各种可能的变化都发生之后为止。而甚至对一个相当大的不能在主存中存放的图像也将其输入/输出(I/O)限制为单次的读/写操作,处理时间近似于使用非约束算子的一次迭代时间。
In multispectral remote sensing data, classifying features using classification techniques produces images with “salt particles” and “peppermint” phenomena that can be removed by logic smoothing but can result in unwanted information loss. With additional connectivity rules to constrain the logic smoothing operator to work on the basic area of the image, you get better results. This paper describes the properties of unconstrained and constrained operators and proposes an algorithm that iteratively applies constraint operators until all possible changes in the image have taken place. Even a fairly large image that can not be stored in main memory also limits its input / output (I / O) to a single read / write operation that takes approximately one iteration of time using a non-constrained operator.