In this paper, we describe an approach for automatic reading of text information from field technician's maps as used in the telecom industry. Given three anchor points as specified by a human operator, the technique isolates the object (cable, or street etc.), identifies all text blocks associated with that object, and converts all information in these text blocks into field data more amenable to future computer manipulation and easy access. The overall algorithm is composed of four main sub-algorithms-contour formation to encompass the object of interest, identification and decomposition of text blocks, identification of individual text blocks by means of HMM based handwriting recognition technique, and finally post-processing of word recognition results by means of field specific knowledge. The success of the algorithm is demonstrated by means of suitable experiments.

A successful handwritten word recognition (HWR) system using a variable duration hidden Markov model (VDHMM) and the path discriminant-HMM (PD-HMM) strategy is easy to implement. The central theme of the paper is to show that if the duration statistics are computed, it could be utilized to implement a model-discriminant-HMM (MD-HMM) approach for better experimental results. The 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.

We describe an MD-HMM (model discriminant HMM) based handwritten word recognition (HWR) system (called NEHMM-nonergodic HMM) whose system parameters are derived from the parameters of the VDHMM (variable duration HMM) system described by Chen, Kundu and Srihari (see IEEE Trans. on Image Proc., vol.4, no.12, p.1675-88, 1995). The new HMM achieves better experimental results by a more efficient utilization of variable duration information. However, more often the problem is 'reliable computation' of duration probabilities given limited databases. A scheme (VSLHMM-variable sequence length HMM) has been presented to avoid the computation of duration probabilities altogether without sacrificing the performance gain of the VDHMM system.

We describe a new block motion estimation scheme by matching similar blocks across the frames under consideration according to their image content. A suitable definition of image content of a block is given. The new algorithm is compared vis-a-vis the conventional algorithm to demonstrate its superiority. Application of this new motion estimation algorithm to video pre-processing is also described.

We introduce a novel approach for handwritten word recognition using multilevel hidden Markov models (MLHMM). The MLHMM is a doubly embedded network of HMMs where each character is modeled by an HMM while a word is modeled by a higher-level HMM. In the character model, we associate the observation with the transition. By introducing the technique called 'tied transition', i.e., the segments which have the same semantic meaning will be 'tied' together, we have successfully built up the character model by an HMM with 4 states, 5 observations (or symbols) and 7 transitions. Thus, as the states are not assigned any semantic meaning, the re-estimation algorithm is applicable. At the character level, the best model is chosen as the recognition result. So, the character model is purely a model discriminant HMM (MD-HMM) based approach. For the word model, on the other hand, both the MD-HMM and the path discriminant HMM (PD-HMM) approaches are used and their respective performances are demonstrated.

Presents an integrated hybrid neural network and hidden Markov model (HMM) classifier that combines the time normalization property of the HMM classifier with the superior discriminative ability of the neural net (NN). Sonar signals display a strong time varying characteristic. Although the neural net has been successful in classifying transient like sonar signals, the success is achieved either by using a bigger net architecture or by incorporating a detection mechanism in the classification procedure. The present authors propose an integrated hybrid HMM and neural net classifier where a left-to-right HMM module is used first. The HMM module segments the observation sequence belonging to every exemplar into a fixed number of states starting from the left. After this segmentation, all the frames belonging to the same state are replaced by one average frame. Thus, every exemplar, irrespective of its time scale variation, is transformed into a fixed number of frames, i.e., a static pattern. The multilayer perceptron (MLP) neural net is then used as the classifier for these time normalized exemplars. For successful modeling and classification, each frame is succinctly represented by a feature vector. Two feature extraction schemes are considered-the first one is based on the FFT power spectral coefficient, and the second one is based on the quadrature mirror filter (QMF) bank based subband decomposition. Finally, some experimental results are provided to demonstrate the superiority of the hybrid integrated classifier.

The JPEG coder has proven to be extremely useful in coding image data. For low bit-rate image coding (0.75 bit or less per pixel), however, the block effect becomes very annoying. The edges also display 'wave-like' appearance. An enhancement algorithm is proposed to enhance the subjective quality of the reconstructed images. First, the pixels of the coded image are classified into three broad categories: (a) pixels belonging to quasi-constant regions where the pixel intensity values vary slowly, (b) pixels belonging to dominant-edge (DE) regions which are characterized by few sharp and dominant edges and (c) pixels belonging to textured regions which are characterized by many small edges and thin-line signals. An adaptive mixture of some well-known spatial filters which uses the pixel labeling information for its adaptation is used as the adaptive optimal spatial filter for image enhancement. Some experimental results are also provided to demonstrate the success of the proposed scheme.

In this paper, we describe an automatic unsupervised texture segmentation scheme using hidden Markov models (HMMs). First, the feature map of the image is formed using Laws' micromasks and directional macromasks. Each pixel in the feature map is represented by a sequence of 4-D feature vectors. The feature sequences belonging to the same texture are modeled as an HMM. Thus, if there are M different textures present in an image, there are M distinct HMMs to be found and trained. Consequently, the unsupervised texture segmentation problem becomes an HMM-based problem, where the appropriate number of HMMs, the associated model parameters, and the discrimination among the HMMs become the foci of our scheme. A two-stage segmentation procedure is used. First, coarse segmentation is used to obtain the approximate number of HMMs and their associated model parameters. Then, fine segmentation is used to accurately estimate the number of HMMs and the model parameters. In these two stages, the critical task of merging the similar HMMs is accomplished by comparing the discrimination information (DI) between the two HMMs against a threshold computed from the distribution of all DI's. A postprocessing stage of multiscale majority filtering is used to further enhance the segmented result. The proposed scheme is highly suitable for pipeline/parallel implementation. Detailed experimental results are reported. These results indicate that the present scheme compares favorably with respect to other successful schemes reported in the literature.

In ocean surveillance, a number of different types of transient signals are observed. These sonar signals are waveforms in one dimension (1-D), and often display an evolutionary pattern over the time scale. The hidden Markov model (HMM) is well-suited to classification of such 1-D signals. Following this intuition, the application of HMM to sonar transient classification is proposed and discussed in this paper. Toward this goal, three different feature vectors based on autoregressive (AR) model, Fourier power spectrum, and wavelet transforms are considered in our work. The neural net (NN) classifier has been successfully used for sonar transient classification. The same set of features as mentioned above is then used with an NN classifier. Some concrete experimental results using "DARPA standard data set I" with HMM and NN classification schemes are presented. Finally, a combined NN/HMM classifier is proposed, and its performance is evaluated with respect to individual classifiers.>

Because of large variation involved in handwritten words, the recognition problem is very difficult. Hidden Markov models (HMM) have been widely and successfully used both in speech and handwriting recognition. Basically, there are two strategies of using HMM: model discriminant HMM (MD-HMM) and path discriminant HMM (PD-HMM). Both of them have their advantages and disadvantages, and are discussed in this paper. Chen, Kundu and Sihari (see Proc. IEEE Int. Conference on Acoust., Speech, Signal Processing, (Minneapolis, Minnesota), p.V.105-108, April 1993) have developed a handwritten word recognition system using continuous density variable duration hidden Markov model (CDVDHMM), which belongs to the PD-HMM strategy. We describe a MD-HMM approach with the statistics derived from the CDVDHMM parameters. Detailed experiments are carried out; and the results using different approaches are compared.>

In this paper, we have developed a handwritten word recognition scheme based on a single contextual hidden Markov model (HMM) incorporated with an adaptive length Viterbi algorithm. This work attempts to extend our earlier HMM scheme for naturally segmented word recognition to cursive and nonsegmented word recognition. The algorithm pre-segments the script into characters and/or fractions of characters, dynamically selects the optimal segmentation points, determines the word length, and recognizes the word according to the maximum path probability. The HMM is on top of, but independent of, script segmentation and character recognition techniques, and therefore can be further improved by incorporating more refined segmentation and character recognition procedure. The experiments have shown promising results.

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.>

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.>

A complete system for the recognition of unconstrained handwritten words using a continuous density variable duration hidden Markov model (CDVDHMM) is described. A new segmentation algorithm based on mathematical morphology is used to translate the 2-D image into a 1-D sequence of sub-character symbols. This sequence of symbols is modeled by the CDVDHMM. Generally, there are two information sources associated with the written text. While the shape information of each character symbol is modeled as a mixture Gaussian distribution, the linguistic knowledge, i.e., constraint, is modeled as a Markov chain. In this context, the variable duration state is used to take care of the segmentation ambiguity among the consecutive characters. Some experimental results are described to demonstrate the success of the proposed scheme.>

A hardware VLSI implementation of the Viterbi algorithm is presented. The Viterbi algorithm is used to solve a word recognition problem using a hidden Markov model. In order to accelerate the speed of computation for real-time word recognition, the inherent parallelism in the recursion step of the algorithm is exploited. Details of the hardware implementation and experimental results are discussed.>

A complete scheme for totally unconstrained handwritten word recognition based on a single contextual hidden Markov model (HMM) is proposed. The scheme includes a morphology- and heuristics-based segmentation algorithm and a modified Viterbi algorithm that searches the (l+1)st globally best path based on the previous l best paths. The results of detailed experiments for which the overall recognition rate is up to 89.4% are reported.>

In the first stage of the proposed scheme the quadrature mirror filter (QMF) bank is used as the wavelet transform to decompose the texture image into subbands. Gray scale transform invariant features 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. During recognition, the unknown texture is matched against all models. The best matched model identifies the texture class. The angles of rotation in the experiments are selected randomly in between -90 degrees and 90 degrees . Up to 95% classification accuracy is reported.>

The authors have developed a handwritten word recognition scheme based on a single contextual, discrete symbol probability hidden Markov model (HMM) incorporated with an adaptive length Viterbi algorithm. This work attempts to extend the earlier HMM scheme for naturally segmented word recognition to cursive and nonsegmented word recognition. The algorithm presegments the script into characters and/or fractions of characters, dynamically selects the correct segmentation points, determines the word length, and recognizes the word according to the maximum path probability. The HMM is on top of, but independent of, script segmentation and character recognition techniques, and therefore leaves room for further improvement. The experiments have shown promising results and directions for further improvement.>

A detail- and structure-preserving smoothing filter is introduced. It is called the combination median filter as it uses directional median, multilevel median, and median filters to smooth different regions of the image. The decision about the region is made by robust Dixon's r-test which is well known in statistics for outlier detection. The threshold value of Dixon's test can be kept constant. As a result, the filtering algorithm operates like a quasi-nonadaptive filter, and no computation of local statistics is involved. Some properties of the filter as well as detailed experimental results that demonstrate its superior performance are presented.

A planar shape-recognition approach is presented which is based on hidden Markov models and autoregressive parameters. This approach segments closed shapes into segments and explores the characteristic relations between consecutive segments to make classification at a finer level. The algorithm can tolerate much shape contour perturbation, and a moderate amount of occlusion. The overall classification scheme is independent of shape orientation. Excellent recognition results have been reported. A distinct advantage of the approach is that the classifier does not have to be trained all over again when a new class of shapes is added.>

A planar shape recognition approach is presented which is based on a hidden Markov model and autoregressive parameters. This approach segments closed shapes into segments and explores the characteristic relations between consecutive segments to make classification at a finer level. The algorithm can tolerate a lot of shape contour perturbation and a moderate amount of occlusion. Also, the overall classification scheme is independent of shape orientation. Excellent recognition results have been reported. A distinct advantage of the approach is that the classifier does not have to be trained all over again when a new class of shapes is added.>

An application of quadrature mirror filter (QMF) bank based subband decomposition to texture analysis is presented. Two-dimensional four bands QMF structure is used and the QMF features are introduced. The low-low band extracts the information of spatial dependence, and the low-high, high-low and high-high bands extract the structural information. This approach has the twin advantages of efficient information extraction and computation reduction. The classification abilities of QMF features are compared to those of Haralick features. The experiments demonstrate that the QMF features have better performance than the Haralick features.>

The authors present a planar shape recognition approach based on the hidden Markov model and autoregressive parameters. This approach segments closed shapes to make classifications at a finer level. The algorithm can tolerate a lot of shape contour perturbation and a moderate amount of occlusion. An orientation scheme is described to make the overall classification insensitive to shape orientation. Excellent recognition results have been reported. A distinct advantage of the approach is that the classifier does not have to be trained again when a new class of shapes is added.>

An algorithm is described for filtering streak noise in corrupted DPCM (differential pulse code modulated) images. In this scheme, the corrupted line is detected using the Wilcoxon-Mann-Whitney nonparametric test. This test mechanism requires no PCM-update multiplexing. The corrupted line is then filtered using a likelihood ratio test and data from the previous line. Typical experimental results are provided.>

A robust edge-detection algorithm which performs equally under a wide variety of noisy situations and a broad range of edges is described. The algorithm is executed in three phases. In phase 1, the step and linear edges are detected from the noise-corrupted image using a statistical classification technique. In phase 2, all the thin-line edges (i.e. which are lines less than two pixels wide) are detected by a supplementary technique since these edges cannot be detected simultaneously with the other step and linear edges. In phase 3, the spurious edge elements are suppressed and the isolated missing edge elements are interpolated using a number of hypothesized edge-segments. Finally some experimental results are provided to illustrate the success of the algorithm.>

The authors consider the problem of filtering images corrupted by non-Gaussian noise using the Kalman filter. Specifically, they have used the reduced updated Kalman filter (RUKF) for the sake of computational savings. They modified the original RUKF to accommodate the non-Gaussian noise using an efficient scheme proposed by them previously (1989). It is shown that the modification is simple and the results are satisfactory.>

The handwritten word recognition problem is modeled in the framework of the hidden Markov model (HMM). The states of HMM are identified with the letters of the alphabet. The optimum symbols are then generated experimentally using 15 different features. Both the first- and second-order HMMs are proposed for the recognition tasks. Using the existing statistical knowledge of English, the calculation scheme of the model parameters are immensely simplified. Once the model is established, the Viterbi algorithm is used to recognize the sequence of letters consisting the word. Some experimental results are also provided indicating the success of the scheme.>

The handwritten script recognition problem is modeled in the framework of the hidden Markov model. For English text, which is the focus of the present research, the states can be identified with the letters of the alphabet, and the optimum symbols can be generated. In order to do so, a quantitative definition of symbols, in terms of features, is required. Fourteen features (some old, some new) are proposed for this task. Using the existing statistical knowledge about the English language, the calculation of the model parameters is immensely simplified. Once the model is established, the Viterbi algorithm is proposed to recognize the single best optimal state sequence, i.e. sequence of letters comprising the word. The modification of the recognition algorithm to accommodate context information is also discussed. Some experimental results are provided indicating the success of the new scheme.>

An algorithm for fitting a conic shape to a set of given data is described. By a modified medial axis transform, described in the paper, the algorithm first determines the orientation of the shape such that the shape could be rotated to align its axis with the X-axis. The standard nonlinear equations of the conic, both closed and open, are then fitted using a linearized least square approach. Some experimental results are provided indicating the convergence and good performance of the algorithm.>

In this paper, a new algorithm for visual shape coding and representation using a powerful theorem on algebraic curve has been presented. The algorithm codes the closed shape, which does never intersect itself, by means of the foci of the closed curve and a distance parameter 'p' such that the product of distances of any point on the shape from the foci is approximately constant and is equal to 'p'. The computation of the foci co-ordinates and 'p' parameter has been presented as a solution to a linearized least-square problem. The reconstruction algorithm is based on straightforward implementation of the theorem. Some experimental results have also been provided indicating the success of the algorithm.

This correspondence describes a new algorithm for extracting edges from natural images. Starting from a simple image model, the algorithm poses the problem of edge extraction as a statistical classifier problem. The algorithm is capable of extracting and detecting edges from images even in the presence of noise. A step by step mathematical derivation of the algorithm reveals the flexibility of the algorithm with pertinent parameters that can be varied for the specific need of the user. Finally, the proposed edge operator is compared to the well-known Marr-Hildreth's edge operator.

In a black-and-white digitized image, pixels corrupted by impulse noise appear as either black dots or white dots. In this paper, the VLSI implementation of two-dimensional generalized mean filtering for the removal of such impulse noises has been presented. The designed chip is capable of filtering each pixel inside the image within 14 clock cycles and is highly suitable for real-time application.