# Last edited on 2012-12-29 21:36:14 by stolfilocal

A module that defines 2D images with float pixels.

Created for Rodrigo Minetto's SnooperText text detector.


History:

  2012-12-15 - J.Stolfi and R.Minetto at UTFPR.
  
    Created module. Not really working.
  
  2012-12-23 - J.Stolfi at UNICAMP.
  
    Extensively re-coded and debugged. In this version, {downsample} and
    {upsample} uses a fixed filter size ({nw = 5}) for all image sizes.
    Therefore the alignment os pixels and point mapping between levels
    are inconsistent. Also, samples near the edges are computed with
    asymmetric filters, leading to artifacts.
    
    Archive: java-FloatImage-js-2012-12-23-214311.tgz
    
  2102-12-26 - J.Stolfi at UNICAMP.
  
    Methods {downsample} and {upsample} recoded again with better handling of
    edge cases. Filter size can be specified by client, and is adjusted to
    match the parity of image size. Point mapping between levels should be
    fixed.
    
    Archive: java-FloatImage-js-2012-12-26-213559.tgz
    
  2102-12-27 - J.Stolfi at UNICAMP.
  
    Added a segmentation module, {FloatImageSegment}. The test program
    also tests segmentation, both positive and negative.
    
    Archive: java-FloatImage-js-2012-12-27-175822.tgz
    
  2012-12-28 - J.Stolfi at UNICAMP.
  
    Fixed {expand-or-crop}; does only integer shifts and does not use
    {BufferedImage} tools, thus aviding quantization errors and 
    loss of detail.  
    
    Added {sign_colorize} to better see the positive and negative parts
    of brightness-normalized images.