Too much irregularity in shapes of intraclass images poises an intrinsic challenge for recognition of handwritten alpha-numeric characters. We propose, in this paper, a perimetric complexity based handwritten numeral recognition that is inspired from gestalt configural superiority effect. A point-wise correspondence is obtained from the matching of shape context using affine transformation. A set of 20 model images are experimentally chosen to obtain composite representations of unknown image which are visually more complex. The proposed numeral recognition system measures this perceptual complexity to recognize an unknown numeral using the Watson's measure. The experiments on ISI Kolkata Odia numeral database show promising results.