Abstract: Similarity between images in image retrieval is measured by computing distances between feature vectors. This paper presents a probabilistic approach and describes two likelihood-based similarity measures for image retrieval. Popular distance measures like the Euclidean distance implicitly assign more more weighting to features with large ranges than those with small ranges. First, we discuss the effects of five feature normalization methods on retrieval performance. Then, we show that the probabilistic methods perform significantly better than geometric approaches like the nearest neighbor rule with city-block or Euclidean distances. They are also more robust to normalization effects and using better models for the features improves the retrieval results compared to making only general assumptions. Experiments on a database of approximately 10000 images show that studying the feature distributions are important and this information should be used in designing feature normalization methods and similarity measures.

Abstract: Causal analysis in multiple equation models often revolves around the evaluation of the effects of an exogenous shift in a structural equation. When taking into account the uncertainty implied by the shape of the likelihood, we argue that how normalization is implemented matters for inferential conclusions around the maximum likelihood (ML) estimates of such effects. We develop a general method that eliminates the distortion of finite-sample inferences about these ML estimates after normalization. We show that our likelihood-preserving normalization always maintains coherent economic interpretations while an arbitrary implementation of normalization can lead to ill-determined inferential results.

Abstract: Statistical mechanics is generalized on the basis of an information theory for inexact or incomplete probability distributions. A parameterized normalization is proposed and leads to a nonextensive entropy. The resulting incomplete statistical mechanics is proved to have the same theoretical characteristics as Tsallis one based on the conventional normalization.

Abstract: We present a measurement of the cluster X-ray luminosity–temperature (L–T) relation out to high redshift (z∼0.8). Combined ROSAT PSPC spectra of 91 galaxy clusters detected in the Wide Angle ROSAT Pointed Survey (WARPS) are simultaneously fitted in redshift and luminosity bins. The resulting temperature and luminosity measurements of these bins, which occupy a region of the high-redshift L–T relation not previously sampled, are compared with existing measurements at low redshift in order to constrain the evolution of the L–T relation. We find the best fit to low-redshift (z 1 keV, to be L∝T3.15±0.06. Our data are consistent with no evolution in the normalization of the L–T relation up to z∼0.8. Combining our results with ASCA measurements taken from the literature, we find η=0.19±0.38 (for Ω0=1, with 1σ errors) where LBol∝(1+z)ηT3.15, or η=0.60±0.38 for Ω0=0.3. This lack of evolution is considered in terms of the entropy-driven evolution of clusters. Further implications for cosmological constraints are also discussed.

Abstract: We show how to characterize compositionally a number of evaluation properties of λ-terms using Intersection Type assignment systems. In particular, we focus on termination properties, such as strong normalization, normalization, head normalization, and weak head normalization. We consider also the persistent versions of such notions. By way of example, we consider also another evaluation property, unrelated to termination, namely reducibility to a closed term. Many of these characterization results are new, to our knowledge, or else they streamline, strengthen, or generalize earlier results in the literature. The completeness parts of the characterizations are proved uniformly for all the properties, using a set-theoretical semantics of intersection types over suitable kinds of stable sets. This technique generalizes Krivine's and Mitchell's methods for strong normalization to other evaluation properties.

Abstract: The harmonics-to-noise ratio (HNR) has been used to quantify the waveform irregularity of voice signals [Yumoto et al., J. Acoust. Soc. Am. 71, 1544-1550 (1982)]. This measure assumes that the signal consists of two components: a harmonic component, which is the common pattern that repeats from cycle-to-cycle, and an additive noise component, which produces the cycle-to-cycle irregularity. It has been shown [J. Qi, J. Acoust. Soc. Am. 92, 2569-2576 (1992)] that a valid computation of the HNR requires a nonlinear time normalization of the cycle wavelets to remove phase differences between them. This paper shows the application of functional data analysis to perform an optimal nonlinear normalization and compute the HNR of voice signals. Results obtained for the same signals using zero-padding, linear normalization, and dynamic programming algorithms are presented for comparison. Functional data analysis offers certain advantages over other approaches: it preserves meaningful features of signal shape, produces differentiable results, and allows flexibility in selecting the optimization criteria for the wavelet alignment. An extension of the technique for the time normalization of simultaneous voice signals (such as acoustic, EGG, and airflow signals) is also shown. The general purpose of this article is to illustrate the potential of functional data analysis as a powerful analytical tool for studying aspects of the voice production process.

Abstract: In this paper, w epropose a subsequence matching algorithm that supports normalization transform in timeseries databases. Normalization transform enables nding sequences with similar uctuation patterns although they are not close to each other before the normalization transform. Application of the existing whole matching algorithm supporting normalization transform to the subsequence matching is feasible, but requires an index for ev ery possible length of the query sequence causing serious overhead on both storage space and update time. The proposed algorithm generates indexes only for a small number of di erent lengths of query sequences. F or subsequence matching it selects the most appropriate index among them. We can obtain better searc h performance by using more indexes. We call our approach index interp olation. We formally pro ve that the proposed algorithm does not cause false dismissal. F or performance evaluation, we have conducted experiments using the indexes for only ve di erent lengths out of the lengths 256 512 of the query sequence. The results show that the proposed algorithm outperforms the sequential scan by up to 14.6 times on the average when the selectivity of the query is 10 .

Abstract: Normalization coefficients in three-dimensional positron emission tomography (3D PET) are affected by parameters such as camera geometry and the design and arrangement of the block detectors. In this work, normalization components for three whole-body 3D-capable tomographs (the GE Advance, the Siemens/CTI 962/HR+ and the Siemens/CTI 951R) are compared by means of a series of scans using uniform cylindrical and rotating line sources. Where applicable, the manufacturers' normalization methods are validated, and it is shown that these methods can be improved upon by using previously published normalization protocols. Those architectural differences between the three tomographs that affect normalization are discussed with a view to drawing more general conclusions about the effect of machine architecture on normalization. The data presented suggest that uniformity of system response becomes easier to achieve as the uniformity of crystal response within the detector block is improved.

Abstract: Unsupervised learning is used to categorize multidimensional data into a number of meaningful classes on the basis of the similarity or correlation between individual samples. In neural-network implementation of various unsupervised algorithms such as principal component analysis, competitive learning or self-organizing map, sample vectors are normalized to equal lengths so that similarity could be easily and efficiently obtained by their dot products. In general, sample vectors span the whole multidimensional feature space and existing normalization methods distort the intrinsic patterns present in the sample set. In this work, a novel method of normalization by mapping the samples to a new space of one more dimension is proposed. The original distribution of the samples in the feature space is shown to be almost preserved in the transformed space. Simple rules are given to map from original space to the normalized space and vice versa.

Abstract: The normalization strategy is popularly used in character recognition to reduce the shape variation. This procedure, however, also gives rise to excessive shape distortion and eliminates some useful information. This paper proposes an aspect ratio adaptive normalization (ARAN) method to overcome the above problems and so as to improve the recognition performance. Experimental results of multilingual character recognition and numeral recognition demonstrate the advantage of ARAN over conventional normalization method.

Abstract: This paper introduces a scheme, which we call the baseline method, to define a measure of term representativeness and measures defined by using the scheme. The representativeness of a term is measured by a normalized characteristic value defined for a set of all documents that contain the term. Normalization is done by comparing the original characteristic value with the characteristic value defined for a randomly chosen document set of the same size. The latter value is estimated by a baseline function obtained by random sampling and logarithmic linear approximation. We found that the distance between the word distribution in a document set and the word distribution in a whole corpus is an effective characteristic value to use for the baseline method. Measures defined by the baseline method have several advantages including that they can be used to compare the representativeness of two terms with very different frequencies, and that they have well-defined threshold values of being representative. In addition, the baseline function for a corpus is robust against differences in corpora; that is, it can be used for normalization in a different corpus that has a different size or is in a different domain.

Abstract: This study explores the extent to which listeners are sensitive to variations in context when listening to Mandarin tones. Specifically, the effects of speaker F0 and speaking rate are evaluated on the perception of a Tone 2-Tone 3 continuum that varied either along a spectral parameter, a temporal parameter, or both. In addition, two groups of listeners were tested, Chinese and American. Results showed that both listener groups compensate for variations in both F0 and speaking rate. However, Chinese and American listeners did not weigh the acoustic cues in the same manner. Results suggest that language background aids in disambiguating phonemic contrasts for Mandarin listeners, but that for English listeners the normalization effects are a consequence of acoustic discriminability. Limitations on perceptual resources allow English listeners to attend to extrinsic information only when intrinsic acoustic differences become more perceptually salient.

Abstract: A comparative analysis is presented for empirical topographic normalization of Landsat Thematic Mapper (TM) data in varied forest and topographic settings. The paired study included rugged areas of Glacier National Park, Montana, Linville Gorge Wilderness, North Carolina, and Green Mountains, Vermont, U.S.A. Empirical models of topographic bias achieved significant corrections in the Montana and Vermont sites. Relative homogeneity of forest structure offset rugged topography in Montana to yield the highest success of normalization. Significant models could not be derived for Linville Gorge. Topographic normalization is most successful when canopy complexity and altitudinal zonation are low to moderate. Atmospheric pollution and geologic control in ridge‐valley alignment are important considerations when undertaking topographic normalization in mountain environments.

Abstract: Methods for normalization of experimental data with experiment-to-experiment variability. The experimental data may include biotechnology data or other data where experiment-to-experiment variability is introduced by an environment used to conduct multiple iterations of the same experiment. Deviations in the experimental data are measured between a central character and data values from multiple indexed data sets. The central character is a value of an ordered comparison determined from the multiple indexed data sets. The central character includes zero-order and low order central characters. Deviations between the central character and the multiple indexed data sets are removed by comparing the central character to the measured deviations from the multiple indexed data sets, thereby reducing deviations between the multiple indexed data sets and thus reducing experiment-to-experiment variability.

Abstract: This paper presents the development of Thai text dependent speaker identification system by applying two feature-feeding approaches. A well-known multilayer perceptron (MLP) network with backpropagation learning algorithm is chosen. It has fast processing time and good performance for pattern recognition problems. But MLP has a limitation in that a network must have a fixed amount of input nodes. Therefore, the linear interpolation time normalization is chosen to adjust the input speech signal into a fixed size of input vector. Furthermore, the windowing technique is developed to avoid the distortion caused by a time normalization process. A fixed size window is sliced through the preprocessed features with fixed amount of overlapping frames. The high identification rate observed in experiments confirms that the developed windowing is suitable for the proposed Thai text-dependent speaker identification system.

Abstract: We repeat the known procedure of the derivation of the set of Proca equations. It is shown that it can be written in various forms. The importance of the normalization is pointed out for the problem of the correct description of spin-1 quantized fields. Finally, the discussion of the so-called Kalb-Ramond field is presented.

Abstract: PROBLEM TO BE SOLVED: To enable retrieval with trade-offs between retrieval precision and retrieval efficiency, and reproducibility and adaptivity at a user's need when a document described in a language having much declension in suffixes of words is retrieved. SOLUTION: This device is provided with a character string information extracting means 26b which extracts a character string (index word) for the index of document data 27a stored in a document DB 27 and a character string (retrieval word) for retrieving in a query text that a user inputs and makes a choice of whether or not the extracted character strings can be normalized and a normalizing means 26c which performs complete normalization for normalizing all normalizable parts when the normalization is performed or incomplete normalization for allowing an unnormalized part to be included and expands the retrieval word into relative notation according to an expansion dictionary 26e. Further, the user is able to specify the normalization and expansion at any time with input indication information.

Abstract: In this contribution a normalization procedure for automatic speech recognition is introduced which aims at reducing speaking rate specific variations of the features of the phonetic classes. A “spurtwise” calculation of normalization factors allows to capture changes of the speaking rate within one utterance. The costsaving implementation using linear interpolation of the original features and a word graph rescoring procedure leads to a moderate increase in computational load compared to the baseline system without speech rate normalization. In addition a two-step procedure which combines vocal tract length normalization (VTLN) and speech rate normalization (SRN) has been developed. Experiments showed, that applying SRN to a VTLN-based recognition system leads to relative reduction in word error rate of 4.2%. This is comparable to the decrease observed when using SRN on a system without VTLN. All in all the combination of VTLN and SRN results in a 15% reduction of word error rate compared to the baseline system.

Abstract: A programmable logic device can be programmed to configure its logic elements to approximate the normalization of probability values used in the operation of logMAP decoders, thereby significantly reducing the amount of logic resources required in the normalization procedure without significantly degrading performance. In the first preferred embodiment, normalization is achieved by approximating the normalization value by calculating an approximate normalization value which is then deducted from all α values in the trellis at any time. This is done by logically ANDing all α input probability values with the NOT of their own MSBs. The resulting outputs are then all bitwise ORed together, the output of which is the approximate normalization value. In another embodiment, the approximate normalization value is calculated using a fixed constant determinable at the outset of the logMAP decoder operation.

Abstract: This paper develops a method that maps an enhanced Entity-Relationship (ER+) schema into a relational schema and normalizes the latter into an inclusion normal form (IN-NF). Unlike classical normalization that concerns individual relations only, IN-NF takes interrelational redundancies into account and characterizes a relational database schema as a whole. The paper formalizes the sources of such interrelational redundancies in ER+ schemas and specifies the method to detect them. Also, we describe briefly a Prolog implementation of the method, developed in the context of a Computed-Aided Software Engineering shell and present a case study.