Automated analysis of ophthalmic angiography sequences usually requires estimating the positions and appearance changes of blood vessels for computer aided diagnosis. Matched filtering is a standard technique for vessel detection on several types of retinal images, but is ineffective on sequences of fluorescein angiograms largely because that the assumption of Gaussian matched filtering can hardly hold in retinal circulation. In this paper we define a non-linear matched filtering using a scaled Gaussian function, which allows the vessels with varying appearance to be detected and the appearance changes to be extracted. The preliminary experimental results obtained from angiographic pairs and images of a SLO sequence are reported.