algorithm - Get interest area of a Histogram (Int Array)

Question

I have an int array full of values (0-255) and I need to extract two indexes.

The indexes to be found will represent the Interesting Area I want to focus on.

The values of this area are always higher than the others but sometimes the difference is very low.

Like this example:

My interest area is this:

My current approach is to get the Max value and Average value. Then I will get:

• the first index greater than "Max - (Max - Average)".
• the last index greater than "Max - (Max - Average)".

But sometimes, like in this case my method gets the Junk part. (The fifth "column", on the right)

Anyone can suggest a better approach?

Note: The interest area has always 4 "columns" like the example image

Answer 1

也许我假设太多了。但是您的数据使 4 座“山”周围看起来有很强的“边缘”。所以你应该看看那些边缘/脊线检测算法组的想法。

我最简单的答案是先尝试阈值。看看这是否有效。

如果没有，可能会应用一维版本的过滤器，例如Sobel 过滤器。它们应该突出显示数据中二阶导数较大的重要点。

Answer 2

您可以对直方图应用简单的脉冲响应滤波器（如 Boxcar）或任何其他平滑方法，以减少高频分量。之后，您可以计算局部最小值和最大值，并在有助于缩小 ROI 的带宽内精确定位各个峰值（每个峰值的最大-最小-最大-最小）。

这是我的一个简单的Boxcar过滤器实现。