This depends on your field. You would not get many particle physicists at a conference with a $1,000 registration fee!
I completely believe you. "Normal" conference fees in Computer Science tend to be in the $300-$500 range. I've mentioned ICPR because there was quite an uproar when the registration fee was announced last year.
There are two things you must keep in mind, though. First is that in Computer Science, conference papers are really, really important. Far more important than in any other field. Journal papers are outdated the moment they are accepted for publication and hopelessly outdated by the time they arrive in your library. So journal papers are reserved for larger breakthroughs which will remain important for a long time.
The second is that top Computer Science conferences have the status of a journal. Their proceedings are widely available, their blind review involves 3-5 reviewers who are likely bigger experts than they would be most journals, lasts more than 3 months and involves a rebuttal phase, just like a journal. So a paper at a conference like ICCV, CVPR, IJCAI, NIPS and the like will do far more for your career than most journal papers. ICPR is not quite that level, but it is just below it, and they are milking it. It's a HUGE conference, with many thousands of visitors, BTW.
I was quite disillusioned when I attended a conference primarily aimed at psychologists, biologists and behavioural scientists. Talks were accepted based on a (barely reviewed) abstract submission and some of the science on display was depressing. Conferences simply don't mean much in many fields, so you're right, it does depend on the field.