2009 |
Štruc, Vitomir; Pavešić, Nikola A comparison of feature normalization techniques for PCA-based palmprint recognition Proceedings Article In: Proceedings of the International Conference on Mathematical Modeling (MATHMOD'09), pp. 2450-2453, Viena, Austria, 2009. Abstract | Links | BibTeX | Tags: biometrics, face verification, feature normalization, normalization, pca, performance evaluation @inproceedings{Mathmod09, Computing user templates (or models) for biometric authentication systems is one of the most crucial steps towards efficient and accurate biometric recognition. The constructed templates should encode user specific information extracted from a sample of a given biometric modality, such as, for example, palmprints, and exhibit a sufficient level of dissimilarity with other templates stored in the systems database. Clearly, the characteristics of the user templates depend on the approach employed for the extraction of biometric features, as well as on the procedure used to normalize the extracted feature vectors. While feature-extraction methods are a well studied topic, for which a vast amount of comparative studies can be found in the literature, normalization techniques lack such studies and are only briefly mentioned in most cases. In this paper we, therefore, apply several normalization techniques to feature vectors extracted from palmprint images by means of principal component analysis (PCA) and perform a comparative analysis on the results. We show that the choice of an appropriate normalization technique greatly influences the performance of the palmprint-based authentication system and can result in error rate reductions of more than 30%. |
Štruc, Vitomir; Pavešić, Nikola Gabor-based kernel-partial-least-squares discrimination features for face recognition Journal Article In: Informatica (Vilnius), vol. 20, no. 1, pp. 115-138, 2009. Abstract | Links | BibTeX | Tags: biometrics, face recogntiion, kernel partial least squares, kpca, lda, pca @article{Inform-Struc_2009, The paper presents a novel method for the extraction of facial features based on the Gabor-wavelet representation of face images and the kernel partial-least-squares discrimination (KPLSD) algorithm. The proposed feature-extraction method, called the Gabor-based kernel partial-least-squares discrimination (GKPLSD), is performed in two consecutive steps. In the first step a set of forty Gabor wavelets is used to extract discriminative and robust facial features, while in the second step the kernel partial-least-squares discrimination technique is used to reduce the dimensionality of the Gabor feature vector and to further enhance its discriminatory power. For optimal performance, the KPLSD-based transformation is implemented using the recently proposed fractional-power-polynomial models. The experimental results based on the XM2VTS and ORL databases show that the GKPLSD approach outperforms feature-extraction methods such as principal component analysis (PCA), linear discriminant analysis (LDA), kernel principal component analysis (KPCA) or generalized discriminant analysis (GDA) as well as combinations of these methods with Gabor representations of the face images. Furthermore, as the KPLSD algorithm is derived from the kernel partial-least-squares regression (KPLSR) model it does not suffer from the small-sample-size problem, which is regularly encountered in the field of face recognition. |
Štruc, Vitomir; Pavešić, Nikola Illumination Invariant Face Recognition by Non-Local Smoothing Proceedings Article In: Biometric ID management and multimodal communication, pp. 1-8, Springer-Verlag, Berlin, Heidelberg, 2009. Abstract | Links | BibTeX | Tags: biometrics, face verification, illumination changes, illumination invariance, illumination normalization, pca, preprocessing @inproceedings{BioID_Multi2009, Existing face recognition techniques struggle with their performance when identities have to be determined (recognized) based on image data captured under challenging illumination conditions. To overcome the susceptibility of the existing techniques to illumination variations numerous normalization techniques have been proposed in the literature. These normalization techniques, however, still exhibit some shortcomings and, thus, offer room for improvement. In this paper we identify the most important weaknesses of the commonly adopted illumination normalization techniques and presents two novel approaches which make use of the recently proposed non-local means algorithm. We assess the performance of the proposed techniques on the YaleB face database and report preliminary results. |
2007 |
Štruc, Vitomir; Pavešić, Nikola Impact of image degradations on the face recognition accuracy Journal Article In: Electrotechnical Review, vol. 74, no. 3, pp. 145-150, 2007. Abstract | Links | BibTeX | Tags: biometrics, face recognition, ica, image degradations, lda, pca @article{EV-Struc_2007, The accuracy of automatic face recognition systems depends on various factors among which robustness and accuracy of the face localization procedure, choice of an appropriate face-feature extraction procedure, as well as use of a suitable matching algorithm are the most important. Current systems perform relatively well whenever test images to be recognized are captured under conditions similar to those of the training images. However, they are not robust enough if there is a difference between test and training images. Changes in image characteristics such as noise, colour depth, background and compression all cause a drop in performance of even the best systems of today. At this point the main question is which image characteristics are the most important in terms of face recognition performance and how they affect the recognition accuracy. This paper addresses these issues and presents performance evaluation (Table 2.) of three popular subspace methods (PCA, LDA and ICA) using ten degraded versions of the XM2VTS face image database [10]. The presented experimental results show the effects of different changes in image characteristics on four score level fusion rules, namely, the maximum, minimum, sum and product rule. All of the feature extraction procedures as well as the fusion strategies are rather insensitive to the presence of noise, JPEG compression, colour depth reduction, and so forth, while on the other hand they all exhibit great sensitivity to degradations such as face occlusion and packet loss simulation |