The purpose of the evaluation was to judge performance of the codecs.  Images of varying complexity in color and texture were used in the evaluation for thorough testing of codec performance.  Compression rate, encoding time and decoding time for the images with varying content was recorded.  Measurement of the "single-ended" encoding and decoding times were accomplished by setting and storing a timer for recording start and stop times.  The time duration was calculated by subtracting the start time from the stop time.  Next, image quality was measured using the two classic metrics of RMSE and SNR.  The following images were used in the evaluation.  Each image has been resized for display purposes.

     
(256 x 256 pixels, 24 bpp)

     
(512 x 512 pixels, 24 bpp)

     
(1024 x 768, 24 bpp)

The table below shows the encoding time of Still Image versus JPEG with the image input of varying content.

The next table shows the decode times for Still Image and JPEG.  Again, the content and dimensions of the input images were varied.

The classic, objective image metrics for image processing are Signal to Noise Ratio (SNR), which is measured in decibels, and Root Mean Square Error (RMSE), which is measured in grey levels.  Both equations compare the original image to the final, decompressed image.  Since SNR is a ratio of signal to noise, a higher number will mean the image has better quality.  Similarly, since RMSE is a measure of error, a lower number will mean better quality.  It also follows that if RMSE is 0, or if SNR has 0 in the denominator, the recovered image is exactly the same as the original image.  Both of these metrics are purely mathematical and are the most widely accepted metrics.

For both the RMSE and SNR curves, in the probable application operating range, Still Image's "Mathematical Image Quality" is better than JPEG for a given compression rate. 

To summarize, the goal in the development of the Still Image codec for image processing is to maximize compression rates and encoding and decoding times while improving image quality.  Still Image encoding and decoding times are faster than JPEG and image quality is better than JPEG at the given compression rates.

Excerpts reprinted with permission of The University of Tennesssee Department of Electrical and Computer Engineering.

MORE>>>