Continue to Site

I need name for a image processing filter description below

Status
Not open for further replies.

wtr

Full Member level 5
Hello,

The filter i'm looking for is a filter that deals with consistency.

Imagine you have power spectrogram, and the corresponding phase difference calculated direction (Angle of arrival +/-180)

So in the image below, the section where a 2x2 block of 45 or -45 (+/- some margin) exist I want the filter to return 1, whereas everywhere else I want 0. This would then be used to correlate with the power mask.

The problem with direct correlation is the number scale.

Example.

545123-1043370
-553037-22384-32120
6134-4523454543-49
-639-44545433
3037573-23-15959-163
`8-5272-80559841
-45-45-4332201-1205
-45-45-43-50373-230

Your example contains 2 such blocks, or 4 such blocks... depending on how you regard the bold-type 2x3 arrangements.

Do you seek a search algorithm?

1) Start a DO loop, examining every square, row by row, column by column.

2). When you see a square which contains Absolute value 43 to 47 (in other words, fuzzy math)...

2a) Set a counter to zero.

2b) Examine its 3 neighbors: right, down, diagonal-right-down. Tally how many contain Absolute value 43 to 47.

2c) If the counter reads 3, then you found a 2x2 group of squares. Set filter output to 1. (Default value=0)

3) Store filter results in a grid with same quantity of rows & columns.

I ended up taking a moving window. What i did was create a list. Then from that list I extracted mathematical properties such as variance.

Using a threshold on the "variance" I was then able to assign the pixel either 1 or 0. This creates a nice mask. This was the consistence filter. Strangely I think an edge detector is the inverse of a consistency filter. I just dont' know what the technical name of this is.

The word variance brings to mind dithering (commonly done to soften the 'jaggies' of a diagonal line drawn on a pixel grid).

Perhaps you wish for your filter to detect an object whose edge was dithered?

Status
Not open for further replies.