Papers - YAMAZAKI Tatsuya

Data-driven segmentation of noisy images modeled with Markov random fields

Tatsuya Yamazaki, Mehdi N. Shirazi, Hideki Noda

Systems and Computers in Japan   24 ( 10 )   62 - 72   1993

　More details

Publishing type：Research paper (scientific journal)  

A data-driven texture-based image segmentation plays an important role in a broad field of application. This paper considers a special case where textures can be modeled by white Gaussian noises. A data-driven segmentation algorithm is developed for images consisting of distinct regions which are corrupted (covered) with region-dependent white Gaussian noise. It is assumed that the number of distinct regions is known whereas the noise parameters are unknown. A recursive algorithm based on the expectation and maximization (EM) method is developed to estimate the parameters of the regional noises. The true scene is considered as a realization of a region process modeled by a Markov random field (MRF) which is observed indirectly. Using the observation image and a provisional estimate of the region process, the region process (MRF's) parameters are estimated by applying the maximum pseudo-likelihood (MPL) method. Then using the estimates of the MRF's parameters, the provisional estimate of the region process is updated by applying the iterated conditional modes (ICM) method. This cycle of parameter estimation and region extraction is continued until it converges. The performance of the proposed algorithm is studied through computer simulation.

Scopus

researchmap

Multi-spectral Image Classification Using Spectral and Spatial Information

Yamazaki Tatsuya, Denis Gingras

ITE Technical Report   17 ( 58 )   7 - 12   1993

　More details

Language：Japanese   Publisher：The Institute of Image Information and Television Engineers  

A new criterion using spectral and spatial information is proposed to classify multi-spectral remote sensing images or textured images. The images are modeled with a hierarchical Markov Random Field(MRF) model, which consists of the observed intensity process and the hidden class label process. The class labels are estimated according to the maximum a posteriori(MAP) criterion, however, some reasonable approximations are done to reduce the computational expenses. Finally a stepwise classification algorithm is derived and simulation and experimental results are shown to verify the validity of this algorithm.

DOI： 10.11485/tvtr.17.58_7

CiNii Article

researchmap

Data-driven restoration of degraded binary MRF images using em method

T. Yamazaki, M. N. Shirazi, H. Noda

IEEE Region 10 Annual International Conference, Proceedings/TENCON   799 - 803   1992

　More details

Publishing type：Research paper (international conference proceedings)  

A data-driven restoration algorithm is developed for binary images degraded nonadditively with flip noises. The true image is assumed to be a realization of a Markov random field (MRF), and the nonadditive flip noises are assumed to be statistically independent and asymmetric. The parameter estimations of the noise and the true image are carried out iteratively in the context of an expectation and maximization (EM) procedure by approximating Baum's auxiliary function of the EM procedure. To approximate Baum's auxiliary function, the estimate of the true image at each EM iteration is used. Simulation examples show the effectiveness of the data-driven restoration algorithm and the parameter estimation procedures.

DOI： 10.1109/TENCON.1992.271860

Scopus

researchmap