Yazar "Afser, Huseyin" seçeneğine göre listele

Listeleniyor 1 - 7 / 7
  • [ X ]
    Öğe
    A Baseline Statistical Method for Robust User-Assisted Multiple Segmentation
    (IEEE-Inst Electrical Electronics Engineers Inc, 2022) Afser, Huseyin
    Recently, several image segmentation methods that welcome and leverage different types of user assistance have been developed. In these methods, the user inputs can be provided by drawing hounding boxes over image objects, drawing scribbles or planting seeds that help to differentiate between image boundaries or by interactively refining the missegmented image regions. Due to the variety in the types and the amounts of these inputs, relative assessment of different segmentation methods becomes difficult. As a possible solution, we propose a simple yet effective, statistical segmentation method that can handle and utilize different input types and amounts. The proposed method is based on robust hypothesis testing, specifically the DGL test, and can be implemented with time complexity that is linear in the number of pixels and quadratic in the number of image regions. Therefore, it is suitable to be used as a baseline method for quick benchmarking and assessing the relative performance improvements of different types of user-assisted segmentation algorithms. We provide a mathematical analysis on the operation of the proposed method, discuss its capabilities and limitations, provide design guidelines and present simulations that validate its operation.
  • [ X ]
    Öğe
    Autocorrelation-based quickest change detection
    (Institute of Electrical and Electronics Engineers Inc., 2020) Afser, Huseyin; Yabaci, Seyhun Barbaros
    We consider the utilization of the autocorrelation information for aiding the quickest detection problem. Specifically, we investigate the problem of quickly detecting a Gaussian source with autocorrelation such that some of its symbols are repeated as cyclic prefixes. Based on the cumulative sum algorithm, we propose a method which takes advantage of this autocorrelation in order to provide performance improvement compared to the classical energy based detection of the uncorrelated source. © 1997-2012 IEEE.
  • [ X ]
    Öğe
    Bayesian Binary Hypothesis Testing under Model Uncertainty
    (Institute of Electrical and Electronics Engineers Inc., 2020) Afser, Huseyin; Yildirim, Ugur
    We consider Bayesian binary hypothesis testing problem when there is only partial knowledge about one of the distributions, while the other distribution is fully known. Specifically, let P1 and P2 be the distributions under two hypothesis, where P2 is known and P1 is unknown. We propose a test and show that if the Chernoff distance between P1 and P2 is known to be larger than ?, an error exponent ?-?, ?>0, can be achieved in the Bayesian setting. If the Chernoff distance between P1 and P2 is not known, but another distribution Q1 known such that l1 distance between P1 and Q1 is known the smaller than ?, then the same test can be applied, and it coincides with the robust hypothesis testing methods existing in the literature. © 2020 IEEE.
  • [ X ]
    Öğe
    Bayesian Binary Hypothesis Testing Under Model Uncertainty
    (IEEE, 2020) Afser, Huseyin; Yildirim, Ugur
    We consider Bayesian binary hypothesis testing problem when there is only partial knowledge about one of the distributions, while the other distribution is fully known. Specifically, let P-1 and P-2 be the distributions under two hypothesis, where P-2 is known and P-1 is unknown. We propose a test and show that if the Chernoff distance between P-1 and P-2 is known to be larger than Phi, an error exponent Phi,- epsilon, epsilon > 0, can be achieved in the Bayesian setting. If the Chernoff distance between P-1 and P-2 is not known, but another distribution Q(1) known such that l(1) distance between P-1 and Q(1) is known the smaller than a, then the same test can be applied, and it coincides with the robust hypothesis testing methods existing in the literature.
  • [ X ]
    Öğe
    Classification With Repeated Observations
    (IEEE-Inst Electrical Electronics Engineers Inc, 2023) Afser, Huseyin; Gyorfi, Laszlo; Walk, Harro
    We study the problem of nonparametric classification with repeated observations. Let X be the d dimensional feature vector and let Y denote the label taking values in {1, . . . , M}. In contrast to usual setup with large sample size n and relatively low dimension d, this letter deals with the situation, when instead of observing a single feature vector X we are given t repeated feature vectors V-1, . . . , V-t. Some simple classification rules are presented such that the conditional error probabilities have exponential rate of convergence as t -> infinity.
  • [ X ]
    Öğe
    Some remarks on Bayesian multiple hypothesis testing
    (Hacettepe Univ, Fac Sci, 2022) Afser, Huseyin
    We consider Bayesian multiple hypothesis problem with independent and identically distributed observations. The classical, Sanov's theorem-based, analysis of the error probability allows one to characterize the best achievable error exponent. However, this analysis does not generalize to the case where the true distributions of the hypothesis are not exact or partially known via some nominal distributions. This problem has practical significance, because the nominal distributions may be quantized versions of the true distributions in a hardware implementation, or they may be estimates of the true distributions obtained from labeled training sequences as in statistical classification. In this paper, we develop a type-based analysis to investigate Bayesian multiple hypothesis testing problem. Our analysis allows one to explicitly calculate the error exponent of a given type and extends the classical analysis. As a generalization of the proposed method, we derive a robust test and obtain its error exponent for the case where the hypothesis distributions are not known but there exist nominal distribution that are close to true distributions in variational distance.
  • [ X ]
    Öğe
    Statistical Classification via Robust Hypothesis Testing: Non-Asymptotic and Simple Bounds
    (IEEE-Inst Electrical Electronics Engineers Inc, 2021) Afser, Huseyin
    We consider Bayesian multiple statistical classification problem in the case where the unknown source distributions are estimated from the labeled training sequences, then the estimates are used as nominal distributions in a robust hypothesis test. Specifically, we employ the DGL test due to Devroye et al. and provide non-asymptotic, exponential upper bounds on the error probability of classification. The proposed upper bounds are simple to evaluate and reveal the effects of the length of the training sequences, the alphabet size and the numbers of hypothesis on the error exponent. The proposed method can also be used for large alphabet sources when the alphabet grows sub-quadratically in the length of the test sequence. The simulations indicate that the performance of the proposed method gets close to that of optimal hypothesis testing as the length of the training sequences increases.