3D image recognition using new set of fractional-order Legendre moments and deep neural networks

作者：

Highlights：

• Novel fractional-order Legendre orthogonal moment invariants (FrOLMIs) for 3D images analysis is derived.

• The local features extraction is studied in order to capture local information from different Region Of Interest (ROI) by adjusting the fractional parameters.

• Fractional-order Legendre orthogonal Moment Invariants in 3D (rotation, scaling & translation) are expressed in terms of fractional-order geometric moment invariants.

• Introduce a novel technique for feature extraction based on fractional-order moment invariants and Deep Neural Networks (DNN) for 3D objects pattern recognition.

• Introduce a new model for 3D objects classification using a set of 3D fractional-order orthogonal moments of Legendre as an input descriptor to a deep neural network structure.

• Providing numerical experiments to demonstrate their efficiency and superiority.

摘要

•Novel fractional-order Legendre orthogonal moment invariants (FrOLMIs) for 3D images analysis is derived.•The local features extraction is studied in order to capture local information from different Region Of Interest (ROI) by adjusting the fractional parameters.•Fractional-order Legendre orthogonal Moment Invariants in 3D (rotation, scaling & translation) are expressed in terms of fractional-order geometric moment invariants.•Introduce a novel technique for feature extraction based on fractional-order moment invariants and Deep Neural Networks (DNN) for 3D objects pattern recognition.•Introduce a new model for 3D objects classification using a set of 3D fractional-order orthogonal moments of Legendre as an input descriptor to a deep neural network structure.•Providing numerical experiments to demonstrate their efficiency and superiority.

论文关键词：Fractional-order orthogonal polynomials,Fractional-order moment invariants,3D image analysis,Global and local features extraction,3D objects recognition,Deep neural networks

论文评审过程：Received 2 November 2020, Revised 15 July 2021, Accepted 10 August 2021, Available online 18 August 2021, Version of Record 24 August 2021.

论文官网地址：https://doi.org/10.1016/j.image.2021.116410