In this paper, 3D models are given by polygonal meshes with the list of vertices and triangular facets. The main difficulty in polygonal meshes is that, the presentation qualities are much dependent on the number of polygons. To solve this problem, an algorithm [14] is used in the pre-processing step to simplify the mesh including the average normal for each triangle in 3D model. In the other hand, the challenge in computing an ideal shape feature is to remain unaffected by scale, rotation and translation transformation. To address this challenge we develop a transformation-invariant descriptor to extract transformation invariant information from the shape indexes [5 and 6] that are relative to a global reference frame presented by the object centroid. The object centroid is extracted directly from the model without any additional processing to the 3D model and during the computation, the amount of the time consumption is smaller.