I have not done a survey of SNP-callers but I am wondering do any of them assign confidence after annotating against dbSNP rather than before?
Intriguing idea. I don't know of such an approach being implemented in a functional, available SNP-caller. However, a couple points come to mind.
First, not all SNPs in dbSNP are of the same standard or confidence.
Two, such a SNP-caller as envisioned in the question would need to take into account population in the sense that a called SNP might have non-zero minor allele frequency (MAF) in south Asians, for example, but zero MAF in Africans (or in one population from Africa). Think of the results of calculations for Neandertal genome in Homo sapiens: ~4% in non-Africans, essentially 0% in Africans.
Three, applying such to admixed populations would be - complex, to say the least.
So, perhaps all I've done is outline some concerns or constraints for algorithm design. Maybe if there is such a SNP-caller out there, these are a couple points to apply to it as a test of its robustness.
A quick literature search found: http://bioinformatics.oxfordjournals.org/content/25/1/6.long
Apparently the new software they have is called SliderII: http://www.bcgsc.ca/platform/bioinfo/software/SliderII
You are able to use known SNPs as priors to inform the SNP calling. I haven't read much into it yet. This will probably go on my huge to-read pile...