In this paper we describe a new framework of feature compensation for robust speech recognition, which is suitable especially for small devices. We introduce Delta-cepstrum Normalization (DCN) that normalizes not only cepstral coefficients, but also their time-derivatives. Cepstral Mean Normalization (CMN) and Mean and Variance Normalization (MVN) are fast and efficient algorithms of environmental adaptation, and have been used widely. In those algorithms, normalization was applied to cepstral coefficients to reduce the irrelevant information from them, but such a normalization was not applied to time-derivative parameters because the reduction of the irrelevant information was not enough. However, Histogram Equalization (HEQ) provides better compensation and can be applied even to the delta and delta-delta cepstra. We investigate various implementation of DCN, and show that we can achieve the best performance when the normalization of the cepstra and the delta cepstra can be mutually interdependent. We evaluate the performance of DCN using speech data recorded by a PDA. DCN provides significant improvements compared to HEQ. It is shown that DCN gives 15% relative word error rate reduction from HEQ. We also examine the possibility of combining Vector Taylor Series (VTS) and DCN. Even though some combinations do not improve the performance of VTS, it is shown that the best combination gives the better performance than VTS alone. Finally, the advantage of DCN in terms of the computation speed is also discussed.

Color histograms are effective for representing color visual features. However, the high dimensionality of feature vectors results in high computational cost. Several transformations, including singular value decomposition (SVD) and principal component analysis (PCA), have been proposed to reduce the dimensionality. In PCA, the dimensionality reduction is achieved by projecting the data to a subspace which contains most of the variance. As a common observation, the PCA basis function with the lowest frquency accounts for the highest variance. Therefore, the PCA subspace may not be the optimal one to represent the intrinsic features of data. In this paper, we apply independent component analysis (ICA) to extract the features in color histograms. PCA is applied to reduce the dimensionality and then ICA is performed on the low-dimensional PCA subspace. The experimental results show that the proposed method (1) significantly reduces the feature dimensions compared with the original color histograms and (2) outperforms other dimension reduction techniques, namely the method based on SVD of quadratic matrix and PCA, in terms of retrieval accuracy.

Color histograms have been considered to be effective for color image indexing and retrieval. However, the histogram only represents the global statistical color information. We propose a new method: A Spatial Weighted Color Histogram (SWCH), for image retrieval. The color space of a color image is partitioned into several color subsets according to hue, saturation and value in HSV color space. Then, the spatial center moment of each subset is calculated as the weight of the corresponding subset. Experiments show that our method is more effective in indexing color image and insensitive to intensity variations.

Chain codes were developed for storing contour information for shape matching. The traditional chain codes are highly susceptible to small perturbations in the contours of the objects. Therefore, traditional chain codes could not be used for image retrieval based on the shape boundaries from the large databases. In this paper, a histogram based chain codes are proposed which could be used for image retrieval. The proposed chain codes are invariant to translation, rotation and scaling transformations, and have high immunity to noise and small perturbations.

In this paper we will present a new color distance measure, that is, angular distance of cumulative histogram. The proposed measure is robust to light variation. We also applied the weitght value to DR, DG, DB according to a Hue histogram of the query image. Moreover, we have compared the measure to previous popular measure that is cumulative L1 distance measure. We show that our method performed more accurate and perceptually relevant result.

This paper proposes an object oriented face region detection and tracking method using range color information. Range segmentation of the objects are obtained from the complicated background using disparity histogram (DH). The facial regions among the range segmented objects are detected using skin-color transform technique that provides a facial region enhanced gray-level image. Computationally efficient matching pixel count (MPC) disparity measure is introduced to enhance the matching accuracy by removing the effect of the unexpected noise in the boundary region. Redundancy operations inherent in the area-based matching operation are removed to enhance the processing speed. For the skin-color transformation, the generalized facial color distribution (GFCD) is modeled by 2D Gaussian function in a normalized color space. Disparity difference histogram (DDH) concept from two consecutive frames is introduced to estimate the range information effectively. Detailed geometrical analysis provides exact variation of range information of moving object. The experimental results show that the proposed algorithm works well in various environments, at a rate of 1 frame per second with 512 480 resolution in general purpose workstation.

A moment-based method is proposed to estimate the illumination change between two images containing affinetransformed objects. The change is linearly modeled with parameters to be estimated by histograms due to its invariance of translation, rotation, and scaling. The parameters can be correctly estimated for an appropriate illumination change by normalizing the moments of the histograms.

In this letter we propose an interpolation technique for low-quality fingerprint images for highly reliable feature extraction. To improve the feature extraction rate, we extract fingerprint features by referring to both the interpolated image obtained by using a directional Laplacian filter and the high-contrast image obtained by using histogram equalization. Experimental results show the applicability of our method.

We propose a recognition method for HMM using a simultaneous generative histogram. Proposed method uses the correlation between two features, which is expressed by a simultaneous generative histogram. Then output probabilities of integrated HMM are conditioned by the codeword of another feature. The proposed method is applied to isolated digit word recognition to confirm its validity.

The authors propose a method of detecting intruders around power line towers using a new image processing technique. With current technology for outdoor imaging, a varitey of factors may lead to erroneous image processing, such as changes of background brightness, rustling of leaves, mist, rain, intrusion of small animals, etc. These problems were solved as follows. With this method, a change of image, which may indicate an intruder, is first detected using a histogram of the brightness difference between a reference image and an observed image. The detected differences are further analyzed to determine whether they represent a human intruder by evaluating a restraint based on the number, the area, the dimensions of the circumscribing rectangle and the center of gravity of the detected portion. Field testing confirmed the method's usefulness, with a successful intruder detection rate of 82%.

This paper proposes a model for learning non-parametric densities using finite-dimensional parametric densities by applying Yamanishi's stochastic analogue of Valiant's probably approximately correct learning model to density estimation. The goal of our learning model is to find, with high probability, a good parametric approximation of the non-parametric target density with sample size and computation time polynomial in parameters of interest. We use a learning algorithm based on the minimum description length (MDL) principle and derive a new general upper bound on the rate of convergence of the MDL estimator to a true non-parametric density. On the basis of this result, we demonstrate polynomial-sample-size learnability of classes of non-parametric densities (defined under some smoothness conditions) in terms of exponential families with polynomial bases, and we prove that under some appropriate conditions, the sample complexity of learning them is bounded as O((1/ε)(2r1)/2r1n(2r1)/2r(1/ε)(1/ε)1n(1/δ) for a smoothness parameter r (a positive integer), where ε and δ are respectively accuracy and confidence parameters. Futher, we demonstrate polynomial-time learnability of classes of non-parametric densities (defined under some smoothness conditions) in terms of histogram densities with equal-length cells, and we prove that under some appropriate condition, the sample complexity of learning them is bounded as O((1/ε)3/21n3/2(1/ε)(1/ε)1n(1/δ)).

This study proposes a new training method for hidden Markov model with separate vector quantization (SVQ-HMM) in speech recognition. The proposed method uses the correlation of two different kinds of features: cepstrum and delta-cepstrum. The correlation is used to decrease the number of reestimation for two features thus the total computation time for training models decreases. The proposed method is applied to Japanese language isolated dgit recognition.