aaalgo / ljpeg

Usage:

0. Install flex with apt-get install flex. Building without flex will fail and pollute the codebase.

1. In your python project

   git clone https://github.com/aaalgo/ljpeg.git
   
   

   Or if you are using git already:

   git submodule add https://github.com/aaalgo/ljpeg.git
   
   

2. Produce the jpeg binary.

   cd jpegdir; make
   
   

A precompiled binary of jpeg is included. In case this step does not work do this:

cd jpegdir;
cp jpeg_static jpeg

3. In your python code:

   from ljpeg import ljpeg
   
   x = ljpeg.read(path)
   

   The loaded is a matrix of type uint16. Typically you want to convert that to float for subsequent processing.

   x = ljpeg.read(path).astype('float')
   

   The LJPEG format sometimes has wrong values for width and height (transposed). One has to read the correct values of width and height from the associating .ics file. Below is a sample snippet for this purpose:

       W = None
       H = None
       # find the shape of image
       for l in open(ics, 'r'):
   	l = l.strip().split(' ')
   	if len(l) < 7:
   	    continue
   	if l[0] == name:
   	    W = int(l[4])
   	    H = int(l[2])
   	    bps = int(l[6])
   	    if bps != 12:
   		logging.warn('BPS != 12: %s' % path)
   	    break
   
       assert W != None
       assert H != None
   
       if W != image.shape[1]:
   	logging.warn('reshape: %s' % path)
   	image = image.reshape((H, W))
   

4. Using ljpeg.py standalone:

   • Convert to TIFF (requires the .ics file in the same directory as LJPEG)

   ./ljpeg.py cases/benigns/benign_01/case0029/C_0029_1.LEFT_CC.LJPEG output.tiff
   

   • Convert to TIFF and verify that no information has been lost

   ./ljpeg.py cases/benigns/benign_01/case0029/C_0029_1.LEFT_CC.LJPEG output.tiff --verify
   

   • Convert to jpeg for visualization with down-sizing scale=0.3 (16-bit TIFF is not good for direct visualization)

   ./ljpeg.py cases/benigns/benign_01/case0029/C_0029_1.LEFT_CC.LJPEG output.jpg --visual --scale 0.3
   

   Note that output file can be any format that's supported by OpenCV (which includes all common types). Most file formats only support 8-bit images, so directly saving into such file formats will cause problems. Add "--visual" to normalize color into 8-bit before saving to such file formats.

The Stanford ljpeg code is in public domain and is therefore OK to be included here. I did minor modification to make the code compile under modern Linux.