Pengecaman Insan Berasaskan Kaedah Profil Sentroid dan Pengelas Rangkaian Neural Buatan

Abstract

Dalam kajian ini, teknik profil sentroid yang berdasarkan pendekatan berasaskan model digunakan bagi tugas pengecaman insan. Kaedah ini dilaksanakan secara mengekstrak ciri–ciri unik perwakilan isyarat gaya lenggang insan serta bukan insan secara automatik dan pasif berasaskan imej pegun. Untuk menilai kekuatan algoritma sarian teknik profil sentroid yang dihasilkan, Rangkaian Neural Buatan (RNB) digunakan sebagai pengelas. Keputusan yang diperolehi membuktikan ciri sarian profil sentroid sesuai digunakan sebagai perwakilan vektor ciri bagi pengelasan insan dengan kadar pengelasan RNB yang dicapai melebihi 98%. Kata kunci: Pengecaman insan; rangkaian neural tiruan; profil sentroid In this study, centroidal profile which is a model based approach is employed for human recognition task. This is done by extracting unique representation of gait features of the subject automatically and passively from static images of human or non human. To evaluate the effectiveness of the generated centroidal profile, Artificial Neural Network (RNB) is used as classifier. Results attained proven that the centroidal profile is appropriate as feature extraction to be used as feature vectors for human shape classification based on classification rate of RNB achieved specifically above 98%. Key words: Human recognition; artificial neural network (ANN); centroidal profile