Performance Comparison of Fourier Transform and its Derivatives as Shape Descriptors for Mango Grading

Mango is a tropical fruit of India which plays a major role in earning foreign currency by export. The export sector of India is paying attention towards it because of its commercial significance. Image has assorted inbuilt features which reflect its content such as color, texture, shape, and spatial relationship features, etc. How to organize and utilize these features effectively in agriculture era is a valuable research topic. Shape is a first quality factor to be considered by consumer while purchasing mango fruit. The purpose of this research work is to explore image processing algorithms and techniques to sort misshapen mango fruits based on their shape features. The developed algorithm would be first step in automated grading and sorting machines in export industries. It can provide a base for fully automatic grading system using computer vision. A shape based mango fruit sorting using computer vision is discussed in this paper. Shape features (Region based and contour based) are designed using Fourier Transform. Wavelet based descriptor is derived from basic Fourier transform to catch local shape details. A two layered radial basis neural network is developed to classify well formed and deformed mango fruits. The experiment results show performance comparison of all Fourier based shape descriptors. Wavelet Fourier Descriptor outperforms region based and contour based Fourier descriptor with classification efficiency of 89. 83%.