Address:

    Room No. 3164, Statistics Discipline, 3rd Academic Building, Khulna University, Khulna-9208, Bangladesh.

    Email:

    Contact:

    +8801737-220964

    Personal Webpage:
    click here

Multi-Class Object Classification Using Diag-Rand HOG; A proposed hybrid approach

An object classification approach based on the hybridization of HOG and LBP was proposed to effectively classify features in complex environments, called Diag-Rand HOG. The efficiency of the proposed algorithm compared to the HOG descriptor was calculated on some well-known datasets like CIFER-10, Fashion-MNIST, and MNIST. We had got an expected outcome compared with other related works by the performance measuring tools like confusion matrix, Precision, Recall, F1-score and some accuracy indicatory graphs. We found the highest accuracy from our proposed model which is 99.2% from MNIST datasets, and 91.15% and 62.15% for Fashion-MNIST and CIFER-10 datasets respectively. Finally, we could hope that, as our feature descriptor along with CNN did a great performance. So, there remained a tremendous chance to work with this feature descriptor in real-life implementations like the renowned classifier and feature extractor.

Details
Role Supervisor
Class / Degree Masters
Students

Abubackar Siddique Sheba

Student ID: 152040

Session: 2018-2019

Start Date January, 2019
End Date March, 2021