Search

A successful handwritten word recognition (HWR) system using Variable Duration Hidden Markov Model (VDHMM) and the PD-HMM strategy is easy to implement. The central theme of this paper is to show that if the duration statistics are computed, it could be utilized to implement an MD-HMM approach for better experimental results. This paper also describes a PD-HMM based HWR system where the duration statistics are not explicitly computed, but results are still comparable to VDHMM based HWR scheme [1]. Index Terms - Variable duration HMM (VDHMM), path discriminant HMM (PD-HMM), model discriminant HMM (MD-HMM), nonergodic HMM (NEHMM), variable sequence length HMM (VSLHMM), adaptive length Viterbi algorithm (ALVA).

A proposed integrated hybrid hidden Markov model and neural network (HMM/NN) classifier is presented to create a unique classifier that would combine the practical and theoretical advantages of both NN and HMM. The experiments included signal description and the selection of training and test samples. Findings indicate that short-time Fourier power spectrum performed better, the integrated HMM/NN classifier's performance is due largely to its ability for better classification and other neural nets other than multilayer perceptrons can be used for classification algorithm.

A novel score function approximation method for filtering non-Gaussian noise environments uses an adaptive normal expansion to efficiently evaluate the score function. Adaptive normal expansions, based on conjugate recentering and the saddle point approach, expand the score function and truncate the higher order terms. The method enables approximation of the score function with some central moments. Experimental results are presented.

A novel continuous density variable duration hidden Markov model (CDVDHMM) recognizes off-line handwritten words. The system utilizes shape information and linguistic knowledge in a suitable manner for the evaluation. A subcharacter segmentation algorithm translates the 2-D image into a 1-D sequential signal, and the CDVDHMM models the 1-D sequences. The use of the model for city name and state name recognition in the postal domain is discussed.

The automatic unsupervised texture segmentation helps in the outline of an image by using Laws' micromasks and directional macromasks. This can be achieved with hidden Markov models (HMMs). The sequential characteristics are modelled as HMMs. The confusion of unnoticed texture segmentation can cause difficulties of the HMMs. The coarse segmentation process enables an approximate count of HMMs and their related model parameters, followed by fine segmentation which gives an exact count. Using the postprocessing step of multiscale majority filtering, the suitable scheme for pipeline or parallel implementation can be achieved.

Because of large variations involved in handwritten words, the recognition problem is very difficult. Hidden Markov Models (HMM) have been widely and successfully used in speech processing and recognition. Recently, HMM has also been used with some success in recognizing handwritten words with presegmented letters. In this paper, a complete scheme for totally unconstrained handwritten word recognition based on a single contextual hidden Markov model type stochastic network is presented. Our scheme includes a morphology and heuristics based segmentation algorithm, a training algorithm that can adapt itself with the changing dictionary, and a modified Viterbi algorithm which searches for the (l + 1)th globally best path based on the previous l best paths. Detailed experiments are carried out and successful recognition results are reported.

In this correspondence, we have presented a rotation and gray scale transform invariant texture recognition scheme using the combination of quadrature mirror filter (QMF) bank and hidden Markov model (HMM). In the first stage, the QMF bank is used as the wavelet transform to decompose the texture image into subbands. The gray scale transform invariant features derived from the statistics based on first-order distribution of gray levels are then extracted from each subband image. In the second stage, the sequence of subbands is modeled as a hidden Markov model (HMM), and one HMM is designed for each class of textures. The HMM is used to exploit the dependence among these subbands, and is able to capture the trend of changes caused by rotation. During recognition, the unknown texture is matched against all the models. The best matched model identifies the texture class. Up to 93.33% classification accuracy is reported.

Proposed modifications to the reduced update Kalman filter to estimate noisy image signals seek to reduce complexity and integrate non-Gaussian noise in the system. Simpler image design can be produced with an efficient local estimator which retains the sharpness and details of the image. Overall filtering performance is shown to be comparable to those of existing Gaussian designs although the proposed method performs better under non-Gaussian conditions.