Inspired by the rotated orientation specific receptive fields of the simple neurons that were discovered by Hubel and Wiesel we describe a multi-layered neural architecture for calculating the local curvature at each point of a planar shape without extracting the underlying contour. Our architecture resembles the visual pathway of primates as we demonstrate how the rotated orientation specific receptive fields of the simple neurons can perform local curvature calculation of the planar shape that is projected on the retina of the eye. We then use the same method to encode planar curvature into the intensity of gray scale images and we demonstrate the effectiveness of this encoding by proposing a shape-based triple correlation invariant image classification scheme. We present experimental results illustrating that by encoding planar curvature into the intensity values we improve the recognition capability of multi layered neural network classifiers without imposing additional complexity to the learning process.