We present in this report a novel approach for shape description based on kernel principal component analysis (KPCA). The strength of this method resides in the affine and particularly scale invariance of KPCA when using a particular kernel referred to as the triangular kernel. Beside this affine invariance, the method provides an effective way to capture the non-linearities in the shape geometry. Experiments conducted on the SQUID database show this affine invariance and the good performances for shape matching and retrieval.