Robust Feature Set Research Articles

A new content-based image retrieval technique using color and texture information

Feature extraction and representation is one of the most important issues in the content-based image retrieval. In this paper, we propose a new content-based image retrieval technique using color and texture information, which achieves higher retrieval efficiency. Firstly, the image is transformed from RGB space to opponent chromaticity space, and the characteristics of the color contents of an image is captured by using Zernike chromaticity distribution moments from the chromaticity space. Secondly, the texture features are extracted using a rotation-invariant and scale-invariant image descriptor in Contourlet domain, which offers an efficient and flexible approximation of early processing in the human visual system. Finally, the combination of the color and texture information provides a robust feature set for color image retrieval. Experimental results show that the proposed color image retrieval is more accurate and efficient in retrieving the user-interested images.

Framework for image retrieval using machine learning and statistical similarity matching techniques

The aim of this study is to take advantage of both shape and texture properties of image to improve the performance of image indexing and retrieval algorithm. Further, a framework for partitioning image into non-overlapping tiles of different sizes, which results in higher retrieval efficiency, is presented. In the new approach, the image is divided into different regions (tiles). Then, the energy and standard deviation of Hartley transform coefficients of each tile, which serve as the local descriptors of texture, are extracted as sub-features. Next, invariant moments of edge image are used to record the shape features. The shape features and combination of sub-features of texture provide a robust feature set for image retrieval. The most similar highest priority (MSHP) principle is used for matching of textural features and Canberra distance is utilised for shape features matching. The retrieved image is the image which has less MSHP and Canberra distance from the query image. The proposed method is evaluated on three different image sets, which contain about 17 000 images. The experimental results indicate that the proposed method achieves higher retrieval accuracy than several previously presented schemes, whereas the computational complexity and processing time of the new method are less than those of other approaches.

Plant cell wall profiling by fast maximum likelihood reconstruction (FMLR) and region-of-interest (ROI) segmentation of solution-state 2D 1H–13C NMR spectra

BackgroundInterest in the detailed lignin and polysaccharide composition of plant cell walls has surged within the past decade partly as a result of biotechnology research aimed at converting biomass to biofuels. High-resolution, solution-state 2D 1H–13C HSQC NMR spectroscopy has proven to be an effective tool for rapid and reproducible fingerprinting of the numerous polysaccharides and lignin components in unfractionated plant cell wall materials, and is therefore a powerful tool for cell wall profiling based on our ability to simultaneously identify and comparatively quantify numerous components within spectra generated in a relatively short time. However, assigning peaks in new spectra, integrating them to provide relative component distributions, and producing color-assigned spectra, are all current bottlenecks to the routine use of such NMR profiling methods.ResultsWe have assembled a high-throughput software platform for plant cell wall profiling that uses spectral deconvolution by Fast Maximum Likelihood Reconstruction (FMLR) to construct a mathematical model of the signals present in a set of related NMR spectra. Combined with a simple region of interest (ROI) table that maps spectral regions to NMR chemical shift assignments of chemical entities, the reconstructions can provide rapid and reproducible fingerprinting of numerous polysaccharide and lignin components in unfractionated cell wall material, including derivation of lignin monomer unit (S:G:H) ratios or the so-called SGH profile. Evidence is presented that ROI-based amplitudes derived from FMLR provide a robust feature set for subsequent multivariate analysis. The utility of this approach is demonstrated on a large transgenic study of Arabidopsis requiring concerted analysis of 91 ROIs (including both assigned and unassigned regions) in the lignin and polysaccharide regions of almost 100 related 2D 1H–13C HSQC spectra.ConclusionsWe show that when a suitable number of replicates are obtained per sample group, the correlated patterns of enriched and depleted cell wall components can be reliably and objectively detected even prior to multivariate analysis. The analysis methodology has been implemented in a publicly-available, cross-platform (Windows/Mac/Linux), web-enabled software application that enables researchers to view and publish detailed annotated spectra in addition to summary reports in simple spreadsheet data formats. The analysis methodology is not limited to studies of plant cell walls but is amenable to any NMR study where ROI segmentation techniques generate meaningful results.Please see Research Article: http://www.biotechnologyforbiofuels.com/content/6/1/46/.

Pedestrians Detection and Part-Based Tracking by Adaptive Segmentation, Neural Network and Kalman Filter

Abstract In this paper authors have proposed a robust pedestrian tracking algorithm for stationary surveillance videos by combining the multiple adaptive segmentation, computationally efficient feature set, Neural Network and Kalman Filter. In the proposed method, human detection has been performed by the use of an adaptive moving average background model with supportive secondary model, which accurately works in changing environment. The parts-based human model has been used with novel and robust feature set for efficient parts-based human recognition. These features are concise, invariant to pose changes and deformation. Pedestrian recognition has been performed by neural network. Visual tracking has been performed by well known Kalman Filter to estimate the target pedestrian position in each next frame. The algorithm is tested on different surveillance videos and results show that proposed pedestrian recognition and tracking method is robust and efficient than similar work done earlier.

Indexing and retrieving in fingerprint databases under structural distortions

This paper presents a new algorithm for fingerprint indexing, which is based on minutia triplets, and it is very tolerant to missing and spurious minutiae. In this sense, a novel representation for fingerprints is proposed by defining a triangle set based on extensions of Delaunay triangulations. Moreover, a set of robust features is used to build indices. Finally, a recovery method based on calculating the recommendation score is introduced, using a new similarity function between geometric transformations. Our proposal was tested on well known databases, showing that it outperforms most of the already reported methods, especially under conditions of distortions.

INVESTIGATING LONG-TERM EFFECTS OF FEATURE EXTRACTION METHODS FOR CONTINUOUS EMG PATTERN CLASSIFICATION

Based on recent advances in modern multifunction myoelectric control devices, a combination of effective feature extraction and classification methods is required to enhance the high classification performance, especially in accuracy viewpoint. However, for realizing practical applications of myoelectric control, the effect of long-term usage or reusability is one of the challenging issues that should be more carefully considered, whereas only a few works have investigated this effect in recent. In this study, the behavior of the state-of-the-art multiple feature extraction methods was investigated with the fluctuating electromyography (EMG) signals recorded during four different days with a large number of trials and subjects. To this end, seven multiple feature sets were compared consisting features based on time domain and time-scale representation. Two major points were emphasized: (1) the optimal robust feature set for continuous (both transient and steady-state signals) EMG pattern classification and (2) the effect of fluctuating EMG signals with feature extraction methods for long-term usage. From the classification results, time domain feature sets yielded better performance than time-scale feature sets. The classification accuracies of the time-domain-feature sets had always achieved above 80% by using linear discriminant analysis (LDA) as a classifier and uncorrelated LDA (ULDA) as a dimensionality reduction, whereas the classification accuracies of the time-scale-feature sets were lower than 70% for the fluctuating EMG signals. The effect of dimensionality reduction for the classification of fluctuating EMG signals was also discussed.

A Fast and Effective Model for Cyclic Analysis and Its Application in Classification

In this paper, a new feature extraction technique for content-based image retrieval is proposed. This method is based on cyclic function (CF) that provides a second-order statistical description in the frequency domain of signals. Then, the CF of each signal is calculated by FFT accumulation method which is a computational efficient algorithm. Features are energy and standard deviation of CF of signals got from image at different regions of bifrequency plane. This scheme shows high performance in both image sets. The image is partitioned into non-overlapping tiles of different sizes. The features drawn from transferred image with proposed new features, CF using first and second moments between the image tiles, serve as local descriptors of texture. This local information is captured for two resolutions and two grid layouts that provide different details of the same image. Shape information is captured in terms of edge. Invariant moments are then used to record the shape features. The combination of the texture features between image and the shape features provide a robust feature set for retrieval. The experimental results show the efficacy of the method. The experimental results are compared with two image sets at previous works and are found to be encouraging.

Unsupervised Content Based Image Retrieval by Combining Visual Features of an Image With A Threshold

Content-based image retrieval (CBIR) uses the visual features of an image such as color, shape and texture to represent and index the image. In a typical content based image retrieval system, a set of images that exhibit visual features similar to that of the query image are returned in response to a query. CLUE (CLUster based image rEtrieval) is a popular CBIR technique that retrieves images by clustering. In this paper, we propose a CBIR system that also retrieves images by clustering just like CLUE. But, the proposed system combines all the features (shape, color, and texture) with a threshold for the purpose. The combination of all the features provides a robust feature set for image retrieval. We evaluated the performance of the proposed system using images of varying size and resolution from image database and compared its performance with that of the other two existing CBIR systems namely UFM and CLUE. We have used four different resolutions of image. Experimentally, we find that the proposed system outperforms the other two existing systems in ecery resolution of image.

Content-based image retrieval by integrating color and texture features

Content-based image retrieval (CBIR) has been an active research topic in the last decade. Feature extraction and representation is one of the most important issues in the CBIR. In this paper, we propose a content-based image retrieval method based on an efficient integration of color and texture features. As its color features, pseudo-Zernike chromaticity distribution moments in opponent chromaticity space are used. As its texture features, rotation-invariant and scale-invariant image descriptor in steerable pyramid domain are adopted, which offers an efficient and flexible approximation of early processing in the human visual system. The integration of color and texture information provides a robust feature set for color image retrieval. Experimental results show that the proposed method yields higher retrieval accuracy than some conventional methods even though its feature vector dimension is not higher than those of the latter for different test DBs.

Robust speaker identification in the presence of car noise

A robust feature set, Teager Energy Operator based Cepstral Coefficients (TEOCC) for speaker identification task is proposed in this paper. Admissible Wavelet Packet (AWP) transform and the Teager Energy Operator (TEO) is used to obtain robust features. The proposed features significantly improve the speaker identification performance (76.25%) compared with the Mel Frequency Cepstral Coefficient (MFCC) features (57%) in the presence of car noise. The performance is evaluated using TIMIT and NOISEX-92 databases. This paper shows that higher-frequency bands also carry more speaker-specific information and the identification rate can be improved without additional processing of the signal to remove noise.

Analysis of Engineered Surfaces for Product Quality Monitoring

Surface texture is the key consideration affecting the function and reliability of engineering components. The engineered textured sur-faces have various quantitative roughness measures that specify the quality of the surface. In industrial product quality monitoring systems, it is increasingly important to devise the automated techniques for evaluation of surface roughness. The proposed method is the novel approach for surface roughness evaluation that implements discrete wavelet transform (DWT) and dual-tree complex wavelet transform (DT-CWT) for analysing surface textures. Further, the method emphasizes selection of effective texture descriptors for classification of surfaces according to their roughness values. The experiments are carried out using surfaces manufactured by the machining processes namely milling, casting, shaping, grinding and blasting. We propose combinations of texture descriptors namely standard deviation, kurtosis, the properties of grey-level co-occurrence matrix (GLCM) and the Canny edge descriptor to form a robust feature set. The Canberra distance metric is used as similarity measure. The algorithm and the results are presented with both DWT and DT-CWT. The feature set comprising standard deviation, kurtosis and GLCM properties gives the correct classification performances of 94.45%, 92.22%, 97.5%, 95%, 95% and 94.38% for milling, casting, shaping, grinding, grit blasting and shot blasting, respectively, with DT-CWT.

An effective image retrieval scheme using color, texture and shape features

In this paper, we present a new and effective color image retrieval scheme for combining all the three i.e. color, texture and shape information, which achieved higher retrieval efficiency. Firstly, the image is predetermined by using fast color quantization algorithm with clusters merging, and then a small number of dominant colors and their percentages can be obtained. Secondly, the spatial texture features are extracted using a steerable filter decomposition, which offers an efficient and flexible approximation of early processing in the human visual system. Thirdly, the pseudo-Zernike moments of an image are used for shape descriptor, which have better features representation capabilities and are more robust to noise than other moment representations. Finally, the combination of the color, texture and shape features provide a robust feature set for image retrieval. Experimental results show that the proposed color image retrieval is more accurate and efficient in retrieving the user-interested images.

Class-level spectral features for emotion recognition

The most common approaches to automatic emotion recognition rely on utterance-level prosodic features. Recent studies have shown that utterance-level statistics of segmental spectral features also contain rich information about expressivity and emotion. In our work we introduce a more fine-grained yet robust set of spectral features: statistics of Mel-Frequency Cepstral Coefficients computed over three phoneme type classes of interest – stressed vowels, unstressed vowels and consonants in the utterance. We investigate performance of our features in the task of speaker-independent emotion recognition using two publicly available datasets. Our experimental results clearly indicate that indeed both the richer set of spectral features and the differentiation between phoneme type classes are beneficial for the task. Classification accuracies are consistently higher for our features compared to prosodic or utterance-level spectral features. Combination of our phoneme class features with prosodic features leads to even further improvement. Given the large number of class-level spectral features, we expected feature selection will improve results even further, but none of several selection methods led to clear gains. Further analyses reveal that spectral features computed from consonant regions of the utterance contain more information about emotion than either stressed or unstressed vowel features. We also explore how emotion recognition accuracy depends on utterance length. We show that, while there is no significant dependence for utterance-level prosodic features, accuracy of emotion recognition using class-level spectral features increases with the utterance length.

A two-stage speech activity detection system considering fractal aspects of prosody

Speech Activity Detectors (SADs) are essential in the noisy environments to provide an acceptable performance in the speech applications, such as speech recognition tasks. In this paper, a two-stage speech activity detection system is presented which at first takes advantage of a voice activity detector to discard pause segments out of the audio signals; this is done even in presence of stationary background noises. In the second stage, the remained segments are classified into speech or non-speech. To find the best feature set in speech/non-speech classification, a large set of robust features are introduced; the optimal subset of these features are chosen by applying a Genetic Algorithm (GA) to the initial feature set. It has been discovered that fractal dimensions of numeric series of prosodic features are the most speech/non-speech differentiating features. Models of the system are trained over a Farsi database, FARSDAT, however, test experiments on the TIMIT English database have been also conducted. Employing the SAD system in conjunction with an ASR system, has been resulted in a relative Word Error Rate (WER) reduction of as high as 28.3%.

Neural localization of acoustic emission sources in ship hulls

In this paper a radial-basis-function neural network is used to localize acoustic emission events in ship hulls. It is shown that using a tiny network configuration and a small set of robust features, selected automatically by the K-means algorithm from a superset of 90 signal parameters, the location of a single event can be classified efficiently into three typical areas found in ship hulls. In simulation experiments, where a stiffened plate model is partially sunk into the water, the localization rate of acoustic emission events in a noise-free environment is 100% using four sensors and only four features per sensor. In the proposed method, the feature set is adapted and estimated automatically in cases of noisy environments. Robust acoustic emission localization rates, greater than 90%, are achieved using less than ten features per sensor in case of additive white Gaussian noise at 0 dB SNR or more.

A Unifying Approach to Moment-Based Shape Orientation and Symmetry Classification

In this paper, the problem of moment-based shape orientation and symmetry classification is jointly considered. A generalization and modification of current state-of-the-art geometric moment-based functions is introduced. The properties of these functions are investigated thoroughly using Fourier series analysis and several observations and closed-form solutions are derived. We demonstrate the connection between the results presented in this work and symmetry detection principles suggested from previous complex moment-based formulations. The proposed analysis offers a unifying framework for shape orientation/symmetry detection. In the context of symmetry classification and matching, the second part of this work presents a frequency domain method, aiming at computing a robust moment-based feature set based on a true polar Fourier representation of image complex gradients and a novel periodicity detection scheme using subspace analysis. The proposed approach removes the requirement for accurate shape centroid estimation, which is the main limitation of moment-based methods, operating in the image spatial domain. The proposed framework demonstrated improved performance, compared to state-of-the-art methods.

Small bowel image classification using cross-co-occurrence matrices on wavelet domain

This paper presents a novel system to compute the automated classification of wireless capsule endoscope images. Classification is achieved by a classical statistical approach, but novel features are extracted from the wavelet domain and they contain both color and texture information. First, a shift-invariant discrete wavelet transform (SIDWT) is computed to ensure that the multiresolution feature extraction scheme is robust to shifts. The SIDWT expands the signal (in a shift-invariant way) over the basis functions which maximize information. Then cross-co-occurrence matrices of wavelet subbands are calculated and used to extract both texture and color information. Canonical discriminant analysis is utilized to reduce the feature space and then a simple 1D classifier with the leave one out method is used to automatically classify normal and abnormal small bowel images. A classification rate of 94.7% is achieved with a database of 75 images (41 normal and 34 abnormal cases). The high success rate could be attributed to the robust feature set which combines multiresolutional color and texture features, with shift, scale and semi-rotational invariance. This result is very promising and the method could be used in a computer-aided diagnosis system or a content-based image retrieval scheme.

Shift-invariant discrete wavelet transform analysis for retinal image classification

This work involves retinal image classification and a novel analysis system was developed. From the compressed domain, the proposed scheme extracts textural features from wavelet coefficients, which describe the relative homogeneity of localized areas of the retinal images. Since the discrete wavelet transform (DWT) is shift-variant, a shift-invariant DWT was explored to ensure that a robust feature set was extracted. To combat the small database size, linear discriminant analysis classification was used with the leave one out method. 38 normal and 48 abnormal (exudates, large drusens, fine drusens, choroidal neovascularization, central vein and artery occlusion, histoplasmosis, arteriosclerotic retinopathy, hemi-central retinal vein occlusion and more) were used and a specificity of 79% and sensitivity of 85.4% were achieved (the average classification rate is 82.2%). The success of the system can be accounted to the highly robust feature set which included translation, scale and semi-rotational, features. Additionally, this technique is database independent since the features were specifically tuned to the pathologies of the human eye.

Significance analysis of qualitative mammographic features, using linear classifiers, neural networks and support vector machines

Advances in modern technologies and computers have enabled digital image processing to become a vital tool in conventional clinical practice, including mammography. However, the core problem of the clinical evaluation of mammographic tumors remains a highly demanding cognitive task. In order for these automated diagnostic systems to perform in levels of sensitivity and specificity similar to that of human experts, it is essential that a robust framework on problem-specific design parameters is formulated. This study is focused on identifying a robust set of clinical features that can be used as the base for designing the input of any computer-aided diagnosis system for automatic mammographic tumor evaluation. A thorough list of clinical features was constructed and the diagnostic value of each feature was verified against current clinical practices by an expert physician. These features were directly or indirectly related to the overall morphological properties of the mammographic tumor or the texture of the fine-scale tissue structures as they appear in the digitized image, while others contained external clinical data of outmost importance, like the patient's age. The entire feature set was used as an annotation list for describing the clinical properties of mammographic tumor cases in a quantitative way, such that subsequent objective analyses were possible. For the purposes of this study, a mammographic image database was created, with complete clinical evaluation descriptions and positive histological verification for each case. All tumors contained in the database were characterized according to the identified clinical features’ set and the resulting dataset was used as input for discrimination and diagnostic value analysis for each one of these features. Specifically, several standard methodologies of statistical significance analysis were employed to create feature rankings according to their discriminating power. Moreover, three different classification models, namely linear classifiers, neural networks and support vector machines, were employed to investigate the true efficiency of each one of them, as well as the overall complexity of the diagnostic task of mammographic tumor characterization. Both the statistical and the classification results have proven the explicit correlation of all the selected features with the final diagnosis, qualifying them as an adequate input base for any type of similar automated diagnosis system. The underlying complexity of the diagnostic task has justified the high value of sophisticated pattern recognition architectures.

Prediction of Temperature in Vacuum Arc Remelting in the Presence of Strong Disturbances

A system for predictive, noninvasive temperature measurement in a broad class of manufacturing processes is presented. The system employs a comprehensive, three-tier signal processing algorithm to achieve high accuracy and reliability in the presence of strong disturbances. Temperature estimates are generated by ‘unstructured, shallow knowledge’ based algorithms. The estimation process employs a set of robust features obtained by signal processing that involves constitutive models representing ‘deep knowledge’ about the process dealt with. The application of these models is combined with ‘structured, shallow knowledge’ based techniques. Results obtained in application to the investment casting of titanium, which employs the vacuum arc remelting (VAR) process, are presented.