Palmprint recognition using gabor wavelet transform
Başlık çevirisi mevcut değil.
- Tez No: 746337
- Danışmanlar: DR. ÖĞR. ÜYESİ AYÇA KURNAZ TÜRKBEN
- Tez Türü: Yüksek Lisans
- Konular: Bilgi ve Belge Yönetimi, Information and Records Management
- Anahtar Kelimeler: Belirtilmemiş.
- Yıl: 2022
- Dil: İngilizce
- Üniversite: Altınbaş Üniversitesi
- Enstitü: Lisansüstü Eğitim Enstitüsü
- Ana Bilim Dalı: Elektrik ve Bilgisayar Mühendisliği Ana Bilim Dalı
- Bilim Dalı: Belirtilmemiş.
- Sayfa Sayısı: 62
Özet
In this thesis, we developed Gabor Wavelet Transform based system for the purpose of palm print recognition with augmentation strategies were evaluated on a subset consisting of five classes. The smaller dataset reduced training time and thus enabled a greater number of experiments in the limited amount of time. It was established that training all five classes on a single multi-class GWT is more effective for augmenting data than generating the samples of each class by a separate network. As the multi-class GWT works with a greater number of training samples and also different classes, it is more likely to capture the underlying structure of a palm pose but also the distinct differences between the action classes. Furthermore, the GWT model with the minimum generator loss had a bigger positive impact on classification accuracy than the model with the best inception accuracy, even though its synthetic sequences looked less realistic. As the best inception accuracy model favors synthetic sequences that are similar to the original training set, the resulting sequences are less likely to contain new information. In contrast to this, the best generator loss takes into account other factors like the continuity between the generated frames and the overall similarity between the generated and the original distribution. By training the classifier solely on synthetic data and then testing it on either generated or real sequences, it could be shown that the created palm sequences contain class-dependent information throughout the sequence and therefore vary for different action classes. The lower visual quality likely comes from the relatively small dataset, the length of the created sequences and the fact that the multi-class GWT probably focuses on the most significant joints. This causes the other joints to either randomly vary or not to change at all. Changes in the architecture or the loss functions of the GWT did lead to better results than the ones achieved by the base architecture with palm print recognition. An SVM-based critic was tested as well. For the implementation in Gabor wavelet transform, it was necessary to remove the gradient penalty from the SVM for palm print recognition as an objective function. While the dynamic design led to the highest improvement in recognition accuracy on the complete dataset, it showed some sign of instability during training and performed worse on the smaller five-class subset.
Özet (Çeviri)
Palm print recognition can be used in many applications such as interactive data analysis or sign detection. Current systems are either expensive, unable to run in real time, or require the user wear devices such as custom gloves. We propose an inexpensive solution for predicting palm prints in real time that uses Gabor Wavelet Transform (GWT) based design. Our system involves training a Support Vector Machine (SVM) classifier with a training, testing threshold of 70% and 30% respectively, and then predicting at a pixel level a naked palm. Our system predicts all pixels at about 10 fps, and is resilient to environment differences in prediction. We also conduct extensive experiment studying the random forest classifier just for comparison purpose with our dedicated work and reveal some interesting properties. It is also shown that the GWT-based data augmentation has been more effective than any other alternative method for interactive data augmentation. In addition to this, the augmentation experiment was repeated on various subsets consisting of different numbers of classes as well as on the complete original palm print sequence dataset. The proposed base architecture increased the classification accuracy on almost all subsets. There was even a slight improvement when the complete dataset was used. The correct identification could be seen as a sign of quality. For the GWT-model with support vector machine (SVM) classifier the best accuracy scored 97.38% was achieved, however the GWT-model with random forest classifier the best accuracy scored 89.03% as given in [29]. The minimum generator loss model therefore outperformed the model with best inception accuracy, which asserts the impression that it is more suitable for the selection of the best GWT model with SVM. The SVM inception network seem to be a suitable choice for the evaluation of the synthetic palm print sample with 3000 features and 12000 samples of images as data augmentation capacities favors data which is similar to the original training set and thus unlikely to contain new information content. The error is only 1.32% with the increasing the number of training samples but the biggest impact in reducing the error can be done by increasing the number of samples.
Benzer Tezler
- Development and evaluation of the authentication systems by using different palm print identificaton methods
Farklı avuç içi tanıma yöntemleri kullanılarak kimlik doğrulama sistemlerinin geliştirilmesi ve değerlendirilmesi
EMEL ÇOŞKUN
Yüksek Lisans
İngilizce
2017
Bilgisayar Mühendisliği Bilimleri-Bilgisayar ve Kontrolİstanbul Kemerburgaz ÜniversitesiDOÇ. DR. OĞUZ BAYAT
- Human identification using palm print images based on deep learning methods and gray wolf optimization algorithm
Derin öğrenme yöntemlerine ve gri kurt optimizasyonu algoritmasına dayalı palmiye baskı görüntüleri kullanarak insan tanımlama
FIRAS HASAN ALI ALSHAKREE
Yüksek Lisans
İngilizce
2022
Bilgisayar Mühendisliği Bilimleri-Bilgisayar ve KontrolÇankırı Karatekin ÜniversitesiElektronik ve Bilgisayar Mühendisliği Ana Bilim Dalı
DR. ÖĞR. ÜYESİ AYHAN AKBAŞ
- Human identification using palm print images
Palm resim kullanarak insan tanımları
MOHAMED R.A ALHASSI
Yüksek Lisans
İngilizce
2017
Bilim ve TeknolojiYıldırım Beyazıt ÜniversitesiElektrik ve Bilgisayar Mühendisliği Ana Bilim Dalı
YRD. DOÇ. DR. ÖZKAN KILIÇ
- Avuç izine dayalı kimliklendirme sistemlerinde kısıtlamasız aya örüntüsü çıkarımı ve uygulaması
Unrestricted palm pattern selection and implementation on palmprint based identification systems
ELİF BAYKAL
Yüksek Lisans
Türkçe
2015
Bilgisayar Mühendisliği Bilimleri-Bilgisayar ve KontrolKaradeniz Teknik ÜniversitesiBilgisayar Mühendisliği Ana Bilim Dalı
DOÇ. DR. MURAT EKİNCİ
- Automatic palmprint recognition
Otomatik avuç içi tanıma
FATHIYA ALI EMHEMED EGREIRA
Yüksek Lisans
İngilizce
2019
Elektrik ve Elektronik Mühendisliğiİstanbul Aydın ÜniversitesiElektrik-Elektronik Ana Bilim Dalı
DR. ÖĞR. ÜYESİ NECİP GÖKHAN KASAPOĞLU