Version 1
: Received: 7 December 2021 / Approved: 8 December 2021 / Online: 8 December 2021 (14:43:39 CET)
How to cite:
Elgharabawy, A.; Prasad, M.; Lin, C.-T. Preference Net: Image Recognition using Ranking Reduction to Classification. Preprints2021, 2021120140. https://doi.org/10.20944/preprints202112.0140.v1
Elgharabawy, A.; Prasad, M.; Lin, C.-T. Preference Net: Image Recognition using Ranking Reduction to Classification. Preprints 2021, 2021120140. https://doi.org/10.20944/preprints202112.0140.v1
Elgharabawy, A.; Prasad, M.; Lin, C.-T. Preference Net: Image Recognition using Ranking Reduction to Classification. Preprints2021, 2021120140. https://doi.org/10.20944/preprints202112.0140.v1
APA Style
Elgharabawy, A., Prasad, M., & Lin, C. T. (2021). Preference Net: Image Recognition using Ranking Reduction to Classification. Preprints. https://doi.org/10.20944/preprints202112.0140.v1
Chicago/Turabian Style
Elgharabawy, A., Mukesh Prasad and Chin-Teng Lin. 2021 "Preference Net: Image Recognition using Ranking Reduction to Classification" Preprints. https://doi.org/10.20944/preprints202112.0140.v1
Abstract
Accuracy and computational cost are the main challenges of deep neural networks in image recognition. This paper proposes anefficient ranking reduction to binary classification approach using a new feed-forward network and feature selection based on rankingthe image pixels. Preference net (PN) is a novel deep ranking learning approach based on Preference Neural Network (PNN), whichuses new ranking objective function and positive smooth staircase (PSS) activation function to accelerate the image pixels’ ranking.PNhas a new type of weighted kernel based on spearman ranking correlation instead of convolution to build the features matrix.ThePNemploys multiple kernels that have different sizes to partial rank image pixels’ in order to find the best features sequence.PNconsists of multiplePNNs’ have shared output layer. Each ranker kernel has a separatePNN. The output results are converted toclassification accuracy using the score function.PNhas promising results comparing to the latest deep learning (DL) networks usingthe weighted average ensemble of eachPNmodels for each kernel on CFAR-10 and Mnist-Fashion datasets in terms of accuracy andless computational cost.
Keywords
Image Recognition; Preference Net
Subject
Computer Science and Mathematics, Computer Vision and Graphics
Copyright:
This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.