Face recognition system in family individuals with content-based image access methods

Image processing, machine learning and deep learning algorithms are used to process data in fields such as facial recognition systems, criminology, security systems. Facial recognition systems can be one of the promising biometric technologies for finding missing individuals, as parents and relatives are more likely to have a face photo of missing individual rather than other biometric methods such as fingerprints or iris. In this thesis, a content-based image retrieval system based on finding missing, kidnapped or wanted individuals is designed and it is aimed to make face estimation with this system. In this context, FEI data set has been provided to find the appropriate machine learning technique to be used in the system. Among Machine Learning techniques on FEI, CelebA and family face dataset; ESA (Evrişimsel sinir Ağları), YGH-DVM (Yönelimli Gradyan Histogramı - Destek Vektör Makineleri) and DSA (Derin Sinir Ağları) methods were used. Evaluating the analysis results of these methods, ESA was determined as the method with the best facial recognition feature, with a 98,86% success rate. Afterwards, the family face reference model was applied to the family face data set, which was created by us and has not been used in any study, and a police sketch was created based on this model. A success rate between 88,77% and 93,90% was obtained by comparing the police sketch of the family reference model of 10 families randomly selected from the family face data set and an individual selected from the family. In addition, a web-based robot image creation program has been prepared.