The bar chart in Matlab is extraordinarily difficult to do anything with that's "out of the box".
To your specific question, the bars are a single object for each column so the width is a fixed single value for the series. The only way I can think to change this without writing something specific would be to use the ability of bar to treat NaN values as placeholders. To do this you would take the range of bins that are wanted to be different and substitute NaN for their values in one column and then do the obverse in a second vector--keep those values in their bins and replace all other with NaN. Then use bar twice, with hold on between the two calls and you'll have two bar plots on the same axes that can set the properties on to differentiate.
>> y=randn(1,1000); % sample data
>> [n,x]=hist(y,20); % get hist data
>> ix=[8:12]; % center bins to accentuate in histogram
>> n=[n; nan(size(n))]; % make second set same size
>> n(2,ix)=n(1,ix); % second is center section values
>> n(1,ix)=nan; % clear first center section (NaN)
>> hB=bar(x,n(1,:)); % outer portion normal bars
>> hold on % get ready to add center
>> hB(2)=bar(x,n(2,:),0.5); % with narrower bars
>>
ADDENDUM Per the comment following IA's alternative, I did a File Exchange search and found <fileexchange/1420-scatterbar3> "SCATTERBAR3(X,Y,Z,WIDTH) draws 3-D bars of height Z at locations X and Y with width WIDTH."
ADDENDUM 2 Following up the discussion there wherein it (finally) dawns on me that 2D histogram is bogus, I think the solution you're looking for is
[n,x]=hist(YourImageArray(:),nBins);
and then follow the above example to emphasize chosen area(s).
