000			02088 a2200157 4500
003			DZ-ElOued
005			20260602171322.0
100	1		_aDJOUADI, Meriem
700	1		_aKHOLLADI, Mohamed-Khireddine
245	0	0	_aImproving location recognition based on multiple images transmitted from a user's phone to aid in way-finding
260			_bUniversite Chahid Hamma Lakhdar d'El-Oued _c2024
500			_aImproving location recognition based on multiple images transmitted from a user's phone to aid in way-finding
520			_aImage geo-localization involves determining the geographical location where a particular image was taken using only its visual information, without relying on any associated metadata. This process often requires advanced techniques in computer vision and machine learning also analyzing features within the image to establish its precise geographical coordinates. Predicting images locations is one of the most challenging and difficult tasks with numerous real-world applications. Lately, significant research has been dedicated to recognizing places and geo-locating images to estimate geographic locations and identify user positions. Recently, Convolutional Neural Networks have shown their capacity to extract significant features in various domains, particularly in computer vision where Deep Learning surpassed previous state-of-the-art methods. Our principal contribution in this dissertation is assessing the use of the extracted features from Pre-trained CNN models and Support Vector Machine (SVM) to improve Street View image classification performance. We assessed our proposed methods on the dataset of Google Street View (GSV), which consists of high-resolution images from the city center as well as the neighboring regions of Pittsburgh, Orlando, and partially Manhattan. The experiment results proved that our proposed approaches significantly improve the accuracy of classification.
650		4	_a/Improving//location//recognition//based//on//multiple//images//transmitted//from//a//user//s//phone//to//aid//in//way//finding/
942			_cTHESIS
999			_c18430 _d18430