honig@ICS.UCI.EDU (David Honig) (01/01/91)
A while ago I asked the Net if anyone was formally evaluating image compression schemes. The answer is: a few. I said I'd post my responses, so here they are. Thanks to all. **************************************** I asked those who responded, What do *you* evaluate exactly? With what goal in mind? What is your backround? (Your address indicates affiliation with an EECS dept; do you work with psychophysicists or manage on your own?) And introduced myself: I'm a grad student studying computational vision, with a backround in computer science and cognitive science/perceptual psych. A few days a week I'm a programmer in the research part of a medical physics & imaging dept of a hospital. There, we have been thinking about running some experiments on the perception of compressed medical images and I was wondering if there were others out there who were willing to share their experiences. By "experiences" I mean experimental methods that they used or pointers to published work where they explain this; also unpublished hints and pitfalls they've found. Technical stuff. E.g., "if you make it brighter, it looks better, so you've gotta normalize..." [when comparing stereos, this is a problem ---the louder one sounds better even if it distorts and shapes the sound...] Did you know that the brightness of a Sony monitor can vary significantly over the face of the tube, but that you don't notice without a photometer? We already have experiences of our own in doing psychophysics on dynamic images, and a "perceptual workstation" (color Sun 4/370, and growing :-) suited to the task. But our experiences were with very simple signal-detection style tasks. Psychophysics traditionally shuns complex, real-world stimuli, due to their uncontrollability. Perhaps there are ways around this problem, eg., by using many images and well-defined (but perhaps realistically complex) tasks? -David Honig **************************************** From: dlluner@mspvmic1.iinus1.ibm.COM To: honig@ICS.UCI.EDU Subject: Psychophysical evaluation of image compression I suspect your conjecture is correct ("Hey Joe -- how's this one look?"). There is a "major" effort underway in the international standards community for the compression of (color) images. The proposed standard (or, almost proposed standard, as the case may be), JPEG, has a number of transmission and/or compression modes based on the DCT (Discrete Cosine Transform -- amusingly engough I just got one of those packets of book RSVP cards from some publisher and, sure enough, there's a book on the DCT now). The DCT is tuneable for varying levels of quality vs. compression. I would hope (though don't expect) there is some work going on in the committee. (Net: I don't know, I'm just pointing to a possibility.) -- David **************************************** From: "Bart M. ter Haar Romeny" <bart@cv.ruu.nl> Subject: Re: anyone doing PSYCHOPHYSICAL testing of image compression??? I know of two groups working in this field, on primarily medical images: prof. George Seeley, Univ. of Arizona, Tucson, Az. is one of the leading experts in this field, prof. Bernie Huang, UCLA, Medical School, Dept. of Radiology and Digital Imaging has done many radiological studies, and built compression routines and hardware himself, for PACS purposes. Good luck. **************************************** From: <haim@hawk.ulowell.edu> Take a look at: ``Performance consequences of two types of stereo picture compression,'' by J. Tselgov, A. Henik, I. Dinstein, and J. Rabany Human Factors, 32(2):173-182, April 1990. --hl **************************************** From: David McLaren <mclaren@tasis.eecs.utas.edu.au> Subject: Re: Psychovisual Compression I am a postgraduate student in the EECS department here. My research involves the compression of images for telecommunication applications. I mainly use the psychophysics of human vision to decide what information I can remove from an image in the compression process (the major characteristics used so far being : spatial frequency response, luminance effects, and spatial masking). No, I don't actually work with any pschophysicists - I get all my information from journals, etc...