Image classification and recognition method to mechanical parts based on fractal dimension

Abstract

Complex mechanical parts have characteristics of irregularity and certain statistical self-similarity, which can be described by fractal dimension. And the values of their fractal dimension can be used as an measurement to classify and recognize the mechanical parts. In addition, the values can guide robots to grab parts. However, the image obtained by a vision system, which contains part images main image and image background will affect the calculation of fractal dimension of main images. In order to solve the problem, an improved differential box-counting method is designed in this paper. The fractal dimension of part images which has been cut and rotated can be calculated using this differential box- counting method. The experimental result shows that the improved differential box-counting method can calculate the fractal dimension of different size-length images, and the values are more stable. The improved method solves the problem that traditional algorithm can only calculate the fractal dimension of image which side length is integer power of 2.