A novel image interpolation algorithm which uses a least-squares approach by means of a spline-like scheme is presented. Such an interpolation can be utilized for image-data compression both in a stand-alone way and as a boundary response in a two-source decomposition scheme. Conventional linear-interpolation approaches are described first. Then, a least-squares bilinear interpolation is proposed. The major advantages of this type of interpolation for image coding are the following: reduction in the tile effect by means of the continuity constraints imposed on the reconstruction and through the noncorrelation of the residual; lower computation complexity: more effective exploitation of interblock correlation compared with traditional block-coding methods; and robustness against noise compared with standard interpolation algorithms. Experimental results show that the efficiency of the proposed algorithm is high, and that higher reconstruction quality can be obtained by an appropriate encoding of the residuals.<<ETX>>