The invention discloses a multi-image joint registration method based on least square estimation. The method solves the problem of improving multi-image registration precision in signal processing ofPS-InSAR and the like, and includes the realization steps of: using Delaunay triangulation to connect all SAR images in an image set, and optimizing the Delaunay triangulation; using an InSAR registration method to estimate registration offsets of already connected SAR image pairs; selecting a reference image in the image set, and selecting certain uniformly-distributed control points on the reference image; using a least square method to estimate registration offsets of all auxiliary images relative to the reference image at each control point; constructing deviation functions of all the auxiliary-image registration offsets, and estimating registration offsets of all the auxiliary images relative to the reference image; and carrying out re-sampling on all the auxiliary images by the offsets to complete multi-image joint registration. The method significantly mitigates time and space decoherence and error propagation effect impacts, improves multi-image joint registration precision, ishigher in a region coincidence degree of the reference image and the auxiliary images at the same pixel, and can be used for signal processing of the PS-InSAR and the like.