IIUM Repository

The feature parallelism model of visual recognition

Hassan, Marwa Yousif and Shuriye, Abdi Omar and Hassan Abdalla Hashim, Aisha and Salami, Momoh Jimoh Eyiomika and Khalifa, Othman Omran (2017) The feature parallelism model of visual recognition. International Journal of Multimedia and Ubiquitous Engineering, 12 (2). pp. 171-186. ISSN 1975-0080

This is the latest version of this item.

[img] PDF - Published Version
Restricted to Repository staff only

Download (2MB) | Request a copy

Abstract

In this work, the Feature Parallelism Model of visual recognition, which addresses the parallel nature of the human brain compared to the hierarchal (serial) brain model, was studied. First, its accuracy rate and training time were compared to those of DeepFace, a leading industry algorithm for face recognition. Both models were trained using ImageNet object recognition dataset. Accuracy rates were almost the same, around 57% top-1 error rate and 33% top-5 error rate. Training time for feature parallelism model has dropped to 21% less than that of Deep Face. Second, we have investigated feature parallelism model under depth, i.e., when adding more layers along the horizontal axis. We have tested the model with 5, 6, 7, and 8 layers respectively; we found that the best results both in terms of accuracy rates and training time were obtained with the sixlayered model. Although the training time enhancement was only a few milliseconds when going from 5 to 6 layers, it has worsened significantly when going from 6 to 7 layers. In fact the training time has tripled, i.e., training time of the 7- layers model is three times of that of the 6- layers model. It continues to worsen by a fewer rate with the 8- layers model. Similarly, accuracy rate was better with the 6- layers model by about 1% of that of the 5- layers model; however, it has worsened by more than 5% whenever we add more layers above six. We consider those results are biologically plausible, as they conform to the biological fact that the cerebral cortex is organized in 6- layers. We’ve concluded that the organization of parallel processing units into 6- layers, either in our brains or in artificial vision systems, may enhance both processing time and accuracy rates.

Item Type: Article (Journal)
Additional Information: 4119/56297
Uncontrolled Keywords: Feature Parallelism Model, Deep learning, Neuroscience, Computer vision
Subjects: T Technology > T Technology (General)
Kulliyyahs/Centres/Divisions/Institutes (Can select more than one option. Press CONTROL button): Kulliyyah of Engineering > Department of Electrical and Computer Engineering
Kulliyyah of Engineering > Department of Mechatronics Engineering
Depositing User: Prof. Dr Othman O. Khalifa
Date Deposited: 09 May 2017 10:54
Last Modified: 01 Feb 2018 01:08
URI: http://irep.iium.edu.my/id/eprint/56655

Available Versions of this Item

  • The feature parallelism model of visual recognition. (deposited 09 May 2017 10:54) [Currently Displayed]

Actions (login required)

View Item View Item

Downloads

Downloads per month over past year