Using multi-scale to measure the similarities between signature sequences is much helpful to obtain high-qualified similarity measures.This paper puts forward a new sequence similarity kernel

called multi-scale sequence spectrum kernel

to measure signature sequence similarity based on shared occurrences of different-scale continuous subsequences.The multi-scale sequence spectrum kernel is optimized through semidefinite program

and used with support vector machine to verify signature directory sequences.The experiments on the benchmark database show the signature verification accuracy has been enhanced

as this approach could automatically optimize the similarity measures with multi-scales

depending on the signature characteristics.