Faculty of Technology

Permanent URI for this community

https://ir.oauife.edu.ng/handle/123456789/6262

Browse

Browsing Faculty of Technology by Author "ALABI, Akeem Adisa"

Now showing 1 - 1 of 1
  • No Thumbnail Available
    Item
    Open Access
    Development of a Segmentation-Based Deformation-Invariant face Recognition model
    (The Department of Computer Science and Engineering, Faculty of Technology, Obafemi Awolowo University., 2022) ALABI, Akeem Adisa
    Facial deformation has been a prominent issue in today’s trend of face recognition being a key product of human’s most frequently observed phenomenon called expression. The fact is that existing models are yet to fully capture aspects of deformation beyond shape and size of the face thus calling for new approaches to improve on present recognition models. This study formulated, implemented and evaluated a segmentation-based model with the view to recognising faces with deformities. A database of face images with different forms of deformation was created by collecting ten (10) different photographs of twenty (20) persons using digital camera and was analysed by discussing their key components. The collected images were grouped into training set and test data. These images were well cropped and then split accordingly through segmentation technique making each of the image features an individual image. The system model was formulated using modified eigenface algorithm incorporated with a three-phase verification key. This gave rise to the introduction of the Aggregated threshold (AT) as against the Uniform Threshold (UT) as the main parameter for the validation of results of comparison during detection and identification of the test images. The system was implemented using Open Computer Vision (i.e. Open CV) with Python programming language. The system evaluation was carried out to determine the role of the Aggregated Threshold as regards to the performance of the model. The evaluation was also extended to determine the behaviour of the system vis-à-vis the change in the pixel evaluation of the set of images. The results obtained showed that the performance of the proposed model outweighed the existing models as far as recognising the test images was concerned. To be precise, all the forty (40) test images were recognised in this model as opposed to the result in the existing model xvi where only 32 out of 40 images were recognized. This represents an increment of 20% accuracy recorded in comparison with the existing model. The use of the Aggregated Threshold through segmentation paved way for harnessing more information from the set of images during training culminating into the successes recorded in this study. It was also observed that the results obtained were similar with little discrepancies in the execution over the range 100x100 down to 50x50 dimensions. In other words, the number of identified test images remained the same with repeated execution of the code. The study concluded that the segmentation procedure introduced in this model gave rise to an enhanced system in the recognition of faces with deformities thereby giving individuals with this problem an opportunity to be recognized by a robust model.