jump to download contact author acknowledgments
Overview
The set was prepared to evaluate the
performance of lossless image compression algorithms for medical and natural
continuous tone grayscale images. The set contains natural continuous tone
grayscale images of various bit depths (up to 16 bits), various sizes (up
to about 4 millions of pixels) and medical images of various modalities
(CR, CT, MR, and US). In the set, image groups were defined, to permit
performance analysis based on average results for the whole group, rather
than on results for single images. The biggest group, called normal, is
for evaluating algorithms’ performance in a typical case, i.e., average
results of compressing images from the normal group may serve as a measure
of algorithms’ average performance for continuous tone gray-scale images.
A collection of smaller groups permits to analyze or compare results with
respect to images’ bit depths, sizes, or medical image modality. The set
contains also non-typical images, which do not belong to the normal group.
To analyze the algorithms’ performance on noisy data special images with
added noise were prepared. To estimate the best-case and the worst-case
performance of algorithms, easily compressible and incompressible pseudo-images
were also generated.
More detailed description of the groups and of the individual images is avaliable here (pdf).
The compression ratios and compression speeds obtained by several algorithms for the set are reported in: Starosolski, R.: Performance evaluation of lossless medical and natural continuous tone image compression algorithms. Proc. SPIE, 2005, Vol. 5959, pp. 116-27, (©).
Other results obtained for the set may be found here and here.
Download the set
The set is publicly available, it may
be downloaded from this page (http://sun.aei.polsl.pl/~rstaros/mednat/index.html).
All the images are stored in the PGM P5 format (grayscale, binary, Big
Endian). The set is split into three parts containing natural, medical,
and other (non-typical) images respectively, each part is contained in
single .zip archive.
natural images (68MB) downloadContact the author
medical images (54MB) download
other images (8MB) download
Acknowledgments
The set was prepared as a part of the
research project Nr 4 T11C 032 24, which was fully supported by the Grant
of the Polish National Research Committee (KBN). Above Grant was carried
out at the Institute of Computer Science, Silesian University of Technology,
in years 2003 and 2004. The author would like to thank people and institutions,
which in helped in obtaining test images and kindly agreed to include the
images in the set. Natural images are scanned photographs by Dr Jacek Szedel
(Silesian University of Technology). Some of the medical images were supplied
by Prof. Ewa Piêtka (Silesian University of Technology), some others are
Philips DICOM Reference Medical Images of Philips Medical Systems (ftp://ftp-wjq.philips.com/medical/interoperability/out/Medical_Images/).
All the remaining medical images are the publicly available DICOM images,
donated by various medical equipment vendors to the RSNA conference (ftp://wuerlim.wustl.edu/pub/dicom/images/version3/).