20.500.12556/DKUM-54261 PREPOZNAVANJE PRIMERKA IZ GENERIČNEGA RAZREDA OBJEKTOV NA OSNOVI DIGITALNIH SLIK IDENTIFYING AN INSTANCE OF A GENERIC CLASS OF OBJECTS BASED ON DIGITAL IMAGES V diplomskem delu je predstavljena metoda za prepoznavanje primerka iz generičnega razreda objektov na osnovi digitalnih slik. Proces prepoznavanja objekta na sliki je ena od hitro razvijajočih se tehnologij. Podajamo pregled že obstoječih metod in postopkov za segmentacijo, strojno učenje ter popis objektov z značilkami. Prav tako je opisana naša implementacija algoritma za prepoznavanje generičnih objektov na sliki. Prepoznavanje primerka (skodelic) smo izvedli s segmentacijo GrubCut, deskriptorji HOG (angl. Histogram of Oriented Gradients), za klasifikacijo primerka pa smo uporabili strojno učenje s stroji SVM (angl. Support Vector Machine). Na testni množici 34 slik z 75 primerki smo izračunali metriko TPR (angl. True Positive Rate) kot 0,92 in metriko PPV (angl. Positive Predicted Value) skoraj 0,96, pri čemer smo uporabili učno množico velikosti 84 slik. In this diploma work is presented a method for identifying an instance of a generic class of objects based on digital images. The process of identifying object in an image is one of the technologies that is rapidly developing nowadays. In this diploma work we present an overview of the existing methods and procedures for segmentation, machine learning and descriptors. In the main part is descibed our implementation of the algorithm for identifying generic object in an image. Identifying objects (cups) we preformed by using GrubCut segmentation, HOG (Histogram of Oriented Gradients) descriptors and for object classification we have used SVM (Support Vector Machine). On the testing dataset with 34 images with 75 instances, we have calculated the TPR (True Positive Rate) metric as 0,92 and PPV (Positive Predictive Value) metric as almost 0,96, where we have used learning dataset with 84 images. Prepoznavanje objektov klasifikacija zaznavanje objektov deskriptor HOG SVM segmentacija Identifying objects classification object detection descriptors HOG SVM segmentation true true false N. Dimovska Slovenski jezik Angleški jezik Diplomsko delo/naloga 2015-08-26 21:00:45 2015-10-14 10:50:07 2022-06-05 03:05:37 2021-05-10 03:35:10 2015 0 [Maribor 0 0000-00-00 NiDoloceno NiDoloceno NiDoloceno 0000-00-00 0000-00-00 0000-00-00 004.932(043.2) 19094294 URN:SI:UM:DK:BACPPRKN UN_Dimovska_Natalija_2015.pdf UN_Dimovska_Natalija_2015.pdf 1 5A009ECB91414D0925928ED376C273F8 31a7c6a90c335b150a19b557d9a9e1b3fcb9b46e59d3ce05ec180406c45d7aff a6bf346c-7c07-11eb-bb7a-00155d0001ca https://dk.um.si/Dokument.php?lang=slv&id=79534 Fakulteta za elektrotehniko, računalništvo in informatiko 0 0 0