如何在嘈杂的图像中找到许多局部最大值？ [英] How can I find many local maxima in a noisy image?

问题描述

如何在2D图像中计算多个最大值，其中最大值不一定都是相同的高度？我发现 imregionalmax（）， imextendedmax（）和 findpeaks（）函数不一定有用，因为它们给出了很多局部最大值，它们实际上只是背景噪声中的最大值。我试过了

How do you go about figuring our multiple max in a 2D image where the max aren't necessarily all the same height? I have found that the imregionalmax(), imextendedmax(), and findpeaks() functions aren't necessarily that helpful because they give many local max that are really just maxes within the background noise. I tried

bw=array > imdilate(array,[1 1 1; 1 0 1; 1 1 1]);

但由于同样的原因，这也是有限的（扩展它使用的矩阵也是如此） ）。

but that also is kind of limited for the same reasons (same thing with expanding the matrix that it uses).

推荐答案

噪声确实是图像分析中的一个问题，你试图找到强度最大值。与图像分析中的任何其他任务一样，您可以通过预处理图像和算法结果的后处理来改善最终结果。

Noise is indeed a problem in image analysis where you try and find intensity maxima. As with any other task in image analysis, you can improve the final results with pre-processing of the image and post-processing of the results of the algorithm.

作为局部最大检测之前的预处理步骤，您可以对图像进行去噪，即过滤图像以抑制一些虚假的最大值（ imfilter 是您可能想要查看的功能）。

As pre-processing step before the local maximum detection, you de-noise the image, i.e. you filter the image to suppress some of the spurious maxima (imfilter is a function you may want to look into).

降噪绝不会消除所有噪音，所以当你进行局部最大值检测时，你仍然会收到许多不必要的最大值。因此，您应用某种启发式来区分好和坏局部最大值。例如，您可以应用强度阈值，低于该阈值可以丢弃所有最大值。

The de-noising never gets rid of all the noise, so when you do the local maximum detection, you still pick up a number of unwanted maxima. Thus, you apply some kind of heuristic to distinguish between 'good' and 'bad' local maxima. For example, you can apply an intensity threshold, below which you discard all maxima.

您可以在此找到一个很好的评论： Smal et al。荧光显微镜中斑点检测方法的定量比较。 IEEE Trans Med Imaging（2010）vol。 29（2）pp.282-301

You find a nice review of this here: Smal et al. Quantitative comparison of spot detection methods in fluorescence microscopy. IEEE Trans Med Imaging (2010) vol. 29 (2) pp. 282-301

这篇关于如何在嘈杂的图像中找到许多局部最大值？的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！