This paper proposes a support vector machine (SVM) algorithm based on probability measure. It employs these probability distributions as embeddings to reproduce kernel Hilbert space (RKHS).In order to reuse many standard kernel-based learning techniques in straightforward fashion

we construct the general form of support vector machine (SVM) called support vector machine based on probability measure (PM - SVM). We set up a virtual database by invariant transformation of the image through the MNIST database

and apply our algorithem to this database

The experimental results demonstrate the effectiveness of this algorithem in rate and efficiency to handwritten digit recognition.