Description of Tests


Test Objectives:

The main questions which the testing of the image compression software sets out to answer are:

1) Do the different PMEM pages perform as expected, i.e. do sky images always compress best under the sky table...?
2) What is the relationship of image entropy to compression ratio and the amount of data loss?
3) How does the GOB length parameter affect compression ratio?
4) What are the advantages/disadvantages of the busy PMEM page with the scaled quantization matrix (PMEM page 1) ?
5) Does the compressor ever perform in unexpected ways?

Test Images:

Testing of the compression software was performed using a set of about 150 Galileo images of Venus and the Moon. These images were obtained in their original VICAR format and have had no previous lossy compression or image processing applied to them. Galileo images were chosen because they are CCD images, and hence are as close as possible to what will be obtained with the Cassini cameras. Voyager images, while their targets may be more suitable, were obtained with a Vidicon camera which has poorer performance than a CCD camera. The images were chosen to encompass a wide range of image entropies and image parameters.

An IDL script was written to determine which images qualify as 'busy', 'sky', and 'atmospheric' by checking which PMEM page gave the best compression ratio with a given image (at a B value of 0). Images compressing best under page 0 are 'busy', those compressing best under page 2 are 'sky', and those compressing best under page 3 are 'atmospheric'. Thumbnail collages of the test images can be found Here. If you are interested in getting copies of the actual raw images I used, please e-mail me at obrien@astrosun.tn.cornell.edu

Test Procedure

An IDL script was written to automate the compression, decompression, and data taking for all of the images. For each group of images (busy, sky and atmospheric), each image was compressed and decompressed with each of the 4 PMEM pages and B values of 0, 1, 2, 3, 5, 10, and 15. A GOB length of zero was used, the NLS bit was set to 0, and optimized decompression was used. The RMS difference between the original image and the decompressed image in each case was found, and is used as a measure of data loss. Also, the compression ratio in each case was read from the compressor log file. Data files were generated for each image type, PMEM page, and B value containing the image names and entropies and the compression ratio and RMS difference value for each image.

Once this data was taken, numerous IDL programs were written to graph and analyze the data. The results appear on the following page.

To determine the dependence of compression ratio on GOB length, three image with a range of entropies were chosen and compressed with different GOB lengths. IDL was used to plot the data and determine the relationship. The results of this also appear on the following page.

Next Page Results of Compression Software Tests
Previous Page Description of Image Entropy
Contents