In this paper we propose a novel method for face recognition on frontal faces. In our method the coarse level shape information of the face is captured by Radon transform as it captures the shape information efficiently. A feature descriptor for each face is computed by applying local binary patterns (LBP) on Radon transform coefficients and computing the histogram of LBP. LBP is used due its computational simplicity and its good texture analysis capabilities. Individual histograms computed on each sub-block of the face image are concatenated in spatial pyramid fashion to attain the complete descriptor. These face descriptors are matched using a distance measure based on pyramid matching kernel(PMK). We evaluated this method using various distance metrics. Experimental results on FERET database shows the significance of this method.