AN ENSEMBLE FEATURE EXTRACTION FOR HAIR BRAID  RECOGNITION

ALEMWORK, MULAT

dc.contributor.author	ALEMWORK, MULAT
dc.date.accessioned	2023-06-19T07:14:42Z
dc.date.available	2023-06-19T07:14:42Z
dc.date.issued	2023-03
dc.identifier.uri	http://ir.bdu.edu.et/handle/123456789/15390
dc.description.abstract	Ethiopia is the owner of different cultures; hair braiding is the most common culture where women braid in different celebrations, especially in the Amhara region. Those different hair braid styles resemble the same but have different arrangements, shapes, patterns, and directions of hair braid. This arrangement of hair braid styles should specify the culture of Gonder, Gojjam, and Wollo hair braiding design. However, due to the different hairstyles of western countries, the pure hair braid style is replaced by those hair styles and loses its cultural identity. There is research conducted on hair segmentation, color classification, and hair style classification. But, there is no research done on Ethiopian women's hair braid recognition by considering the direction, shape, and position. This research work focused on designing an ensemble feature extraction for hair braid recognition of Ethiopian women especially in the Amhara region of Gonder, Gojjam, and Wollo based on braid design. The image of the hair braid was captured from different places around Gonder, Debre Tabor, and Bahir dar city. The image size we used by taking three image sample sizes that are 360 x 360 pixels, 224x224 pixels, and 512x512 pixels. We used Convolutional Neural Network (CNN) for an extracted deep feature and handcrafted Histogram Oriented Gradient (HOG) for extracted shape, position, and direction features of hair braid design. We have also compared CNN with AlexNet, VGGNet, and LeNet with SVM classifiers and we achieved a highest accuracy of 93.89 %. In this study, we have to use the Support Vector Machine (SVM) algorithm for the classification of hair braid design into the specified class. Accordingly, HOG feature extraction, CNN feature extraction, and combined feature vector are experimented with by applied SVM classifier, in which 90 %, 93.89 %, and 95 % accuracy were achieved respectively. Keywords: Hair braids style, CNN, HOG, Feature extraction, SVM	en_US
dc.language.iso	en_US	en_US
dc.subject	Computing	en_US
dc.title	AN ENSEMBLE FEATURE EXTRACTION FOR HAIR BRAID RECOGNITION	en_US
dc.type	Thesis	en_US