Imlib and other Software

imlib-src.tar.gz contains both C++ image libraries and applications, and the newmat11 matrix libraries. See the README file for some of the contents. The distribution is updated sporadically but is fairly up to date. Platforms: windows/Unix/Linux. I have recently (dec-2005) had problems with the new gcc compiler series (3.2 and later), I have tried to change the code accordingly but there might be some link errors.

A deblurring example: This is a confocal microscope image depicting chromosomes I picked up from the internet. The first image (256.jpg) shows the raw image, the second image (256_deconv.jpg) shows the image after 10 iterations of Lucy-Richardson (LR) deblurring. In this case the Poisson noise model is particularly useful. In general LR deblurring seems to work well for astronomical images and for pictures with heavy deblurring. For low SNR images more refined regularisation models must be used. In the deconv.cpp program I use here there is a "a-trous" wavelet pyramid noise control built in. This means that wavelet coefficients below a threshold value are discarded in the wavelet reconstruction. Here is a review of deconvolution for optical astronomical images. Here is the script (windows bat file) that does the convolution and which uses the "deconv.exe" program in Imlib. Since I didn't know the deblurring function I assumed a Gaussian with stdv=1.1, a value found after some trial and error.

256 256_proc

You can also calculate local orientation with imlib. The method used here is to derive the orientation vector of the inertia matrix in the spatial domain (J₂₂ - J₁₂, 2J₁₂) J_pq = B(D_p·D_q), where B is a blurring operator and D is a one-dimensional (Gaussian) derivative operator, and the "dot" is point-wise multiplication. The color coding is derived from the phase of the orientation vector arctan(2J₁₂/(J₂₂ - J₁₁), where blue is +π/2 and red -π/2. The gif-animation below depicts the newly crew-cut head of yours truly, where the local orientation has been superposed on an edge map, pruned with non-maximal suppression (using the program "edgeori.cpp" in the imlib distribution). Nice books on low-level image processing can be found here.

head

There is a doxyfile in the distribution that will generate a html-code documentation

A simple C++ spectral line fitting program adapted for supernova spectra. This is a self contained distribution that also contains the newmat11 matrix library. It does spectral line and Planck curve fitting using a Levenberg-Marquardt procedure with analytical expressions for the jacobian. The spectrum is smoothed using a Savitzky-Golay filter to preserve spectral shapes. The programs automatically finds local maxima and fits a skewed gaussian a * x * exp(-b x^2) to the smoothed data. The output is presented as textfiles and as gnuplot (.dem) files for quick inspection. An example run is shown below. There are three examples of spectra in the distribution and data is generated as text files using a tcsh script and the user can inspect data using gnuplot files.

super1 super2 super3 super4

Platforms: Linux/Unix. Updated 2004-12-21, should now also work for the cygwin environment.