[Insight-users] image io

[Ziv Yaniv]

Hi all,
After using the itk image IO for a while I have one question and one 
wish-list request:

1. When reading image data is there a way to read data without 
specifying the pixel type in advance?
   typedef itk::Image<int,3> ImageType;
   typedef itk::ImageReader<ImageType> ReaderType;

The ITK toolkit makes the assumption that I know the pixel type in 
advance which is not always the case. This primarily happens when I read 
DICOM images. The toolkit itself loads the data correctly into memory as 
it knows what the pixel type is after reading the DICOM header. Once the 
data is read it is cast to the pixel type that I had provided (e.g. int, 
see above). As I'm not sure what the pixel type is I will usually use 
the largest possible primitive (double), which is a waste of memory. Is 
there an elegant way to query the file and then create a correct image 
type and reader?

2. When writing 2D images (jpg/png/bmp) the writers check that the image 
is 2D and that the pixel type is valid (e.g. unsigned int), what they 
don't check is that the image is isotropic which is what these image 
file formats assume. Please add this as an additional condition for 
successful image writing (took me a while to understand why my extracted 
slices were so distorted).

                           best
                                  Ziv

P.S. From the itkJPEGImageIO.h (itk 2.8):

  /** Determine the file type. Returns true if this ImageIO can read the
   * file specified. */
  virtual bool CanWriteFile(const char*);

  Should be "... can write the ...", the joys of cut and paste.