Which test should I use?

Will your quantified images result in a single continuous random variable per test subject or some other scale of quantitation? If you end up with a single quantitative outcome on a continuous scale per subject then you can use an ANOVA to test for a difference between the means of the four groups. Depending on how many samples you have in each group you'll want to check the data vs the assumptions of an ANOVA. If the ANOVA is significant then move on to a Duncan's t with your control group as the control. You don't want to perform multiple t tests without controlling the Type I error rate. The ANOVA followed by a Duncan's t will achieve that for you. If you can't reasonably assume that the assumptions of the model are being met you'll need to move on to a nonparametric test method.

I think each subject will have a continuous random variable...protein level is continuous? I will only have 5-6 samples per group though.

What would the Duncan's t-test help to achieve?

Actually for the control, I may have only 3 samples instead of 5.

Sorry, I meant Dunnett's not Duncan's.

I'm not sure how Image J works. Does it give you a single quantified outcome for the entire blot? Does it differentiate between the banding regions? I'd have to know a lot more about your outcomes before I could advise you how best to proceed. 5 samples per group is EXTREMELY small for anything except an exact test. Can you categorize the outcomes and do a Fisher's Exact Test on the cells?

Dunnett's t is a pairwise analysis of the means of the groups against a control. If your interest is in how the three groups compare individually to the control group then use Dunnett's t. If you're interested in how each of the groups compare individually to each other and the control then use Duncan's or Tukey's tests for pairwise comparisons.