Matlab, comparing array using if statement

Question

I have searched the net trying to find an answer to this problem I have.

I have an array much like the following

A = [2 4 6 8 ; 3 5 7 9 ; 1 4 6 9]

row median = [ 5 6 5 ]
col median = [ 2 4 6 9 ]

From these values I want to create a median map. So I have created the array

MedianMap = int8(zeros(MAX_ROWS, MAX_COLS))

Within this array I want to assign three different values: 1, 0, -1. So the median map output will be of the same size of array 'A':

• if the value is greater than both the row and column median a "1" is assigned to the median map
• if the value is less than both the row and column median a "-1" is assigned to the median map
• otherwise a 0?

How can I traverse through every row and column in the "A" array and relate it to its respective column and row median?

I have written the code in C code and it was sucessful, however just struggling in Matlab.

Answer 1

Here is how I would do it:

1. Create logical indices for each condition (element greater/less than row/col median)
2. Use the logical indices to update MedianMap.

In code:

[xMedian, yMedian] = meshgrid(col_median, row_median);

isRowHigh = (A > yMedian);
isColHigh = (A > xMedian);

isRowLow = (A < yMedian);
isColLow = (A < xMedian);

MedianMap(isRowHigh & isColHigh) = 1;
MedianMap(isRowLow & isColLow) = -1;

Notes:

• meshgrid expands row_median and col_median into arrays of the same size as A
• A > yMedian returns a matrix of the same size as A containing the boolean results of comparing every element of A with the corresponding element of xMedian.
• isRowHigh & isColHigh performs an element-wise AND of the boolean matrices
• MedianMap(L), where L is a logical index (boolean matrix), selects the elements of MedianMap corresponding to the elements of L which are true.

Answer 2

Here's how I would do that:

MedianMap = ...
    ( bsxfun(@gt,A,col_median) & bsxfun(@gt,A,row_median.') ) - ...
    ( bsxfun(@lt,A,col_median) & bsxfun(@lt,A,row_median.') );

This one is multi-threaded (suited for much larger problems) and doesn't have any of the temporaries involved in the other answers (much smaller peak memory footprint).

It's not very pretty though :) So if better readability is what you're after, use either meshgrid as in BrianL's answer, or repmat:

Col_median = repmat(col_median, size(A,1),1);
Row_median = repmat(row_median.', 1, size(A,2));

MedianMap = ...
    ( A > Col_median & A > Row_median ) - ... 
    ( A < Col_median & A < Row_median ); 

or multiplication by a ones-matrix as Rasman did:

Col_median = ones(size(A,1),1) * col_median;
Row_median = row_median.' * ones(1,size(A,2));

MedianMap = ...
    ( A > Col_median & A > Row_median ) - ... 
    ( A < Col_median & A < Row_median ); 

Answer 3

MedianMap  = (A > Rmedian'*ones(1,4))+ ( A > ones(3,1)*Cmedian) -1