Thanks for pointing out a bug in HDRREAD. The .hdr format (which uses RGBE encoding) has a special condition when all of the samples for one pixel are less than 10E-38; those pixels should get mapped to (0,0,0,0), and they do in HDRWRITE. But I neglected to put similar code in HDRREAD to convert (0,0,0,0) to (0,0,0) as the spec requires.

I will try my best to get that fix into the next version of the Image Processing Toolbox, with a workaround available even sooner.

BTW, I think you might notice that the values of t are quite a bit smaller than 0.5740. When I run the code I see 0.5740E-41.

Jeff

I’ve been experimenting with some HDR algorithms, and I wanted to now use the .hdr file format to pass some intermediate results from MATLAB to some C++ code I have.

However, I’m finding some weird rounding errors when writing/ reading hdr images with MATLAB, even just writing them out then reading them back in. Specifically, very small values get rounded to zero when I write them to then read them back from an hdr file. More generally, why are values changing a lot when doing this? Does the hdr format require some re-scaling or rounding of values?

Here’s a simple example.

q=zeros(10,10,3);
q(:,:,:)=0;
hdrwrite(q,'test.hdr');
t=hdrread('test.hdr')

When I run this code, all the entries for t are 0.5740. With actual images, what I’ll tend to see is a number like 3.042e-5 turn into a zero. The result is that in the HDR weighting matrix sum, the smallest numbers (which would give the biggest final composite values since) turn to zeros resulting in NANs in the final HDR image, so the total dynamic range is much smaller than it should be.

Thanks.

The MAKEHDR function doesn’t currently take camera response curves into account at all. Consequently, there’s nothing to save for later use. (I know, it’s a rather naïve approach.)

You might be interested in this article from Peter Web that shows another method of HDR generation using MATLAB: http://www.mathworks.com/company/newsletters/digest/2008/july/hdri.html He does compute camera response curves, which could be useful to your work.

I am currently doing masters research related to High Dynamic Range Imaging. In particular, I am working on ghost removal. This usually requires being able to apply the camera response curve in different manners in different parts of the image.

Can you recommend a way to get the camera response curve and save it so it can be used manually later and applied to various parts of the image?

I don’t see this functionality currently in the Matlab IP toolbox’s HDR functions, but it seems like an obvious thing to want. Thanks.

>> I = imread(‘cameraman.tif’);
tform = maketform(‘affine’,[1 0 0; .5 1 0; 0 0 1]);
J = imtransform(I,tform);
imshow(I), figure, imshow(J)
??? Error using ==> imtransform>parse_inputs at 440
XData and YData could not be automatically determined.Try specifying XData and YData explicitly in the call to IMTRANSFORM.

Error in ==> imtransform at 262
args = parse_inputs(varargin{:});

Now of course I can enter xdata and ydata to map the old image to the new one but why am I getting this error on an example?

Thanks,
John

If I may, I have another question: I am making a ‘custom’ spatial transformation and I am using pol2cart and cart2pol as the forward and inverse functions. I get the following error message:

??? Undefined function or method ‘atan2′ for input arguments of type ‘struct’.

Error in ==> cart2pol at 22
th = atan2(y,x);

Error in ==> maketform>inv_composite at 584
U = feval(t.tdata(i).inverse_fcn, U, t.tdata(i));

Error in ==> images\private\tform at 56
X = feval( t.(f.fwd_fcn), U, t );

Error in ==> tforminv at 68
varargout = tform(‘inv’, nargout, varargin{:});

Error in ==> tformarray at 241
M = tforminv(G,T);

Error in ==> imtransform at 273
B = tformarray(args.A, args.tform, args.resampler, tdims_a, tdims_b, …

Error in ==> rphi2xy_imtransf at 76
Z = imtransform(absd_chiqui,TFORM, ‘UData’,udata,’VData’,vdata,’XData’,xdata,’YData’,ydata);

My code is like this:
dims_IN = 2;
dims_OUT = 2;
FRWD_FCN = @pol2cart;
INV_FCN = @cart2pol;
TDATA = [];
TFORM = maketform(‘custom’,dims_IN, dims_OUT, FRWD_FCN,INV_FCN,TDATA);
udata = [range_start range_end];
vdata = [-angle_w/2 angle_w/2];
xdata = [range_start range_end];
ydata = [range_end*sin(angle_w/2) -range_end*sin(angle_w/2)];
Z = imtransform(absd_chiqui,TFORM, ‘UData’,udata,’VData’,vdata,’XData’,xdata,’YData’,ydata);

Initially I did not write any UData, VDAta, XData or YData and I got this error message:
XData and YData could not be automatically determined.Try specifying XData and YData explicitly in the call to IMTRANSFORM.

So I guess my question is: isn´t it possible to use such functions for a custom transformation, or is my problem in the TFORM definition?

Thank you again for your time,

Pia