Mesh processing is a wide area which consists of several approaches, heavily specific for each task. We present a novel approach to mesh processing using the Local Projections method. The described method makes benefit of wide variety of image processing algorithms, very similar to 3D tasks, and implements a conversion mechanism to make possible use of these processes on polygonal models. We also designed, implemented and evaluated the curvature approximation algorithm to test and compare the proposed method usability on real and artificial data. Use of the proposed method brings significant benefits especially to noised mesh analysis.