Features For Image Classification Research Articles

Automatic segmentation of breast lesions for interaction in ultrasonic computer-aided diagnosis

Breast cancer is one of the most commonly diagnosed cancer types among women. Sonography has been regarded as an important imaging modality for diagnosis of breast lesions. Due to the speckle and the variance in shape and appearance of sonographic lesions, fully automatic segmentation of the breast tumor regions still remains a challenging task. In this paper, we propose an automatic interaction scheme based on an object recognition method to segment the lesions in breast ultrasound images. In this scheme, a 2D ultrasound image is firstly filtered with a total-variation model to reduce the speckle noise. A robust graph-based segmentation method is then used to segment the image into a number of sub-regions. An object recognition method incorporating the procedures of image feature extraction, feature selection and classification is proposed to automatically identify the regions which are associated with breast tumors. Finally, an active contour model is used to refine the contours of the regions that are recognized as tumors. This scheme is validated on a database of 46 breast ultrasound images with diagnosed tumors. The experimental results show that our scheme can segment the breast ultrasound images automatically, indicating its good performance in real applications.

Image registration algorithm using Mexican hat function-based operator and grouped feature matching strategy.

Feature detection and matching are crucial for robust and reliable image registration. Although many methods have been developed, they commonly focus on only one class of image features. The methods that combine two or more classes of features are still novel and significant. In this work, methods for feature detection and matching are proposed. A Mexican hat function-based operator is used for image feature detection, including the local area detection and the feature point detection. For the local area detection, we use the Mexican hat operator for image filtering, and then the zero-crossing points are extracted and merged into the area borders. For the feature point detection, the Mexican hat operator is performed in scale space to get the key points. After the feature detection, an image registration is achieved by using the two classes of image features. The feature points are grouped according to a standardized region that contains correspondence to the local area, precise registration is achieved eventually by the grouped points. An image transformation matrix is estimated by the feature points in a region and then the best one is chosen through competition of a set of the transformation matrices. This strategy has been named the Grouped Sample Consensus (GCS). The GCS has also ability for removing the outliers effectively. The experimental results show that the proposed algorithm has high registration accuracy and small computational volume.

Formation features for improving the quality of medical diagnosis based on the discriminant analysis methods

The computer diagnostic system of eye diseases is considered. To improve the quality of diagnostics we propose an algorithm for the informative features formation, using methods of discriminant analysis. The method for receiving an informtiveness estimation is described. The research confirming the efficiency of the formed features for classification of images of an fundus was conducted by means of classification by support vector machine. The algorithm possesses a sufficient level of universality and may be applied to increase the informtiveness of any feature set.

Feature Extraction and Classification of Images Based on Corner Invariant Moments

Image feature extraction and classification is increasingly important in all sectors of the images system management. Aiming at the problems that applying Hu invariant moments to extract image feature computes large and too dimensions, this paper presented Harris corner invariant moments algorithm. This algorithm only calculates corner coordinates, so can reduce the corner matching dimensions. Combined with the SVM (Support Vector Machine) classification method, we conducted a classification for a large number of images, and the result shows that using this algorithm to extract invariant moments and classifying can achieve better classification accuracy.

Diabetic Rethinopathy Screening by Bright Lesions Extraction from Fundus Images

Abstract Retinal images are nowadays widely used to diagnose many diseases, for example diabetic retinopathy. In our work, we propose the algorithm for the screening application, which identifies the patients with such severe diabetic complication as diabetic retinopathy is, in early phase. In the application we use the patient’s fundus photography without any additional examination by an ophtalmologist. After this screening identification, other examination methods should be considered and the patient’s follow-up by a doctor is necessary. Our application is composed of three principal modules including fundus image preprocessing, feature extraction and feature classification. Image preprocessing module has the role of luminance normalization, contrast enhancement and optical disk masking. Feature extraction module includes two stages: bright lesions candidates localization and candidates feature extraction. We selected 16 statistical and structural features. For feature classification, we use multilayer perceptron (MLP) with one hidden layer. We classify images into two classes. Feature classification efficiency is about 93 percent.

改进的LFP算法在白细胞图像纹理特征提取与识别中的应用

Leukocyte image feature extraction and classification are studied to improve the correct recognition rate of leukocyte image.For cell texture feature extraction,leukocyte image texture features are extracted by using the improved local fuzzy pattern,and the texture feature extraction method based on local fussy pattern(LFP)was proposed by making the threshold parameter fuzzy in local binary pattern(LBP).The algorithm introduces in uniform pattern to make the extracted feature dimension decrease to 10 with rotation invariance.The classification of 100CellAtlas' s white blood cells images was tested with a support vector machine combination classifier established by a directed acyclic graph method.Experimental results show that:the improved local fuzzy pattern algorithm simplifies texture feature quantity to realize"discard the false and retain the true".The leukocyte image classification and recognition with noise exhibits excellent performance,so that the extracted features have better Robustness.And it has a short running time,high efficiency,leukocyte correct recognition rate is up to 93%.Improved support vector machine classifier shows efficient classification effect,and has better characteristics to small sample analysis.

Two-dimensional bilinear preserving projections for image feature extraction and classification

Two-dimensional locality preserving projections (2DLPP) was recently proposed to extract features directly from image matrices based on locality preserving criterion. A significant drawback of 2DLPP is that it only works on one direction (left or right) to reduce the dimensionality of the image matrices and thus too many coefficients are needed for image representation in low-dimensional subspace. In this paper, we propose a novel method called two-dimensional bilinear preserving projections (2DBPP) for image feature extraction. We generalized the image-based (2D-based) feature extraction techniques into bilinear cases, in which 2DLPP is a special case of our proposed method. In order to obtain the bilinear projections, we proposed an iteration method by solving the corresponding generalized eigen-equations. Moreover, analyses show that 2DBPP has stronger locality preserving abilities than 2DLPP. By using the label information and defining different local neighborhood graphs, the proposed framework is further extended to supervised case. Experiments on three databases show that 2DBPP and its supervised extension are superior to some other image-based state-of-the-art techniques.

Automatic Tortuosity Classification Using Machine Learning Approach

Retinopathy of Prematurity (ROP) is a vital cause of vision loss in premature infants, but early detection of its symptoms enables timely treatment and prevents blindness. Tortuosity is the major indicator of ROP that can potentially be automatically quantified. In this paper, which focuses on automatic tortuosity quantification and classification in images from infants at risk of ROP, we present a series of experiments on preprocessing, feature extraction, image feature selection and classification using nearest neighbor classifier. Fisher linear Discriminant analysis is used as a feature selection algorithm. We observe that the best feature set is a combination of two features: tortuosity as estimated based on combination of curvature of improved chain code and number of inflections and tortuosity as measured by inflection count metric. Accuracy, sensitivity and specificity are used as performance measures for the classifier. The results are validated against the judgments of expert ophthalmologists. The overall accuracy, sensitivity and specificity achieved on the best feature set are 95%, 95.65% and 96.74% respectively.

Supervised and Unsupervised Parallel Subspace Learning for Large-Scale Image Recognition

Subspace learning is an effective and widely used image feature extraction and classification technique. However, for the large-scale image recognition issue in real-world applications, many subspace learning methods often suffer from large computational burden. In order to reduce the computational time and improve the recognition performance of subspace learning technique under this situation, we introduce the idea of parallel computing which can reduce the time complexity by splitting the original task into several subtasks. We develop a parallel subspace learning framework. In this framework, we first divide the sample set into several subsets by designing two random data division strategies that are equal data division and unequal data division. These two strategies correspond to equal and unequal computational abilities of nodes under parallel computing environment. Next, we calculate projection vectors from each subset in parallel. The graph embedding technique is employed to provide a general formulation for parallel feature extraction. After combining the extracted features from all nodes, we present a unified criterion to select most distinctive features for classification. Under the developed framework, we separately propose supervised and unsupervised parallel subspace learning approaches, which are called parallel linear discriminant analysis (PLDA) and parallel locality preserving projection (PLPP). PLDA selects features with the largest Fisher scores by estimating the weighted and unweighted sample scatter, while PLPP selects features with the smallest Laplacian scores by constructing a whole affinity matrix. Theoretically, we analyze the time complexities of proposed approaches and provide the fundamental supports for applying random division strategies. In the experiments, we establish two real parallel computing environments and employ four public image and video databases as the test data. Experimental results demonstrate that the proposed approaches outperform several related supervised and unsupervised subspace learning methods, and significantly reduce the computational time.

Computer-assisted lip diagnosis on traditional Chinese medicine using multi-class support vector machines

BackgroundIn Traditional Chinese Medicine (TCM), the lip diagnosis is an important diagnostic method which has a long history and is applied widely. The lip color of a person is considered as a symptom to reflect the physical conditions of organs in the body. However, the traditional diagnostic approach is mainly based on observation by doctor’s nude eyes, which is non-quantitative and subjective. The non-quantitative approach largely depends on the doctor’s experience and influences accurate the diagnosis and treatment in TCM. Developing new quantification methods to identify the exact syndrome based on the lip diagnosis of TCM becomes urgent and important. In this paper, we design a computer-assisted classification model to provide an automatic and quantitative approach for the diagnosis of TCM based on the lip images.MethodsA computer-assisted classification method is designed and applied for syndrome diagnosis based on the lip images. Our purpose is to classify the lip images into four groups: deep-red, red, purple and pale. The proposed scheme consists of four steps including the lip image preprocessing, image feature extraction, feature selection and classification. The extracted 84 features contain the lip color space component, texture and moment features. Feature subset selection is performed by using SVM-RFE (Support Vector Machine with recursive feature elimination), mRMR (minimum Redundancy Maximum Relevance) and IG (information gain). Classification model is constructed based on the collected lip image features using multi-class SVM and Weighted multi-class SVM (WSVM). In addition, we compare SVM with k-nearest neighbor (kNN) algorithm, Multiple Asymmetric Partial Least Squares Classifier (MAPLSC) and Naïve Bayes for the diagnosis performance comparison. All displayed faces image have obtained consent from the participants.ResultsA total of 257 lip images are collected for the modeling of lip diagnosis in TCM. The feature selection method SVM-RFE selects 9 important features which are composed of 5 color component features, 3 texture features and 1 moment feature. SVM, MAPLSC, Naïve Bayes, kNN showed better classification results based on the 9 selected features than the results obtained from all the 84 features. The total classification accuracy of the five methods is 84%, 81%, 79% and 81%, 77%, respectively. So SVM achieves the best classification accuracy. The classification accuracy of SVM is 81%, 71%, 89% and 86% on Deep-red, Pale Purple, Red and lip image models, respectively. While with the feature selection algorithm mRMR and IG, the total classification accuracy of WSVM achieves the best classification accuracy. Therefore, the results show that the system can achieve best classification accuracy combined with SVM classifiers and SVM-REF feature selection algorithm.ConclusionsA diagnostic system is proposed, which firstly segments the lip from the original facial image based on the Chan-Vese level set model and Otsu method, then extracts three kinds of features (color space features, Haralick co-occurrence features and Zernike moment features) on the lip image. Meanwhile, SVM-REF is adopted to select the optimal features. Finally, SVM is applied to classify the four classes. Besides, we also compare different feature selection algorithms and classifiers to verify our system. So the developed automatic and quantitative diagnosis system of TCM is effective to distinguish four lip image classes: Deep-red, Purple, Red and Pale. This study puts forward a new method and idea for the quantitative examination on lip diagnosis of TCM, as well as provides a template for objective diagnosis in TCM.

Image-based continental shelf habitat mapping using novel automated data extraction techniques

We automatically mapped the distribution of temperate continental shelf rocky reef habitats with a high degree of confidence using colour, texture, rugosity and patchiness features extracted from images in conjunction with machine-learning algorithms. This demonstrated the potential of novel automation routines to expedite the complex and time-consuming process of seabed mapping. The random forests ensemble classifier outperformed other tree-based algorithms and also offered some valuable built-in model performance assessment tools. Habitat prediction using random forests performed most accurately when all 26 image-derived predictors were included in the model. This produced an overall habitat prediction accuracy of 84% (with a kappa statistic of 0.793) when compared to nine distinct habitat classes assigned by a human annotator. Predictions for three habitat classes were all within the 95% confidence intervals, indicating close agreement between observed and predicted habitat classes. Misclassified images were mostly unevenly, partially or insufficiently illuminated and came mostly from rugged terrains and during the autonomous underwater vehicle's obstacle avoidance manoeuvres. The remaining misclassified images were wrongly or inconsistently labelled by the human annotator. This study demonstrates the suitability of autonomous underwater vehicles to effectively sample benthic habitats and the ability of automated data handling techniques to extract and reliably process large volumes of seabed image data. Our methods for image feature extraction and classification are repeatable, cost-effective and well suited to studies that require non-extractive and/or co-located sampling, e.g. in marine reserves and for monitoring the recovery from physical impacts, e.g. from bottom fishing activities. The methods are transferable to other continental shelf areas and to other disciplines such as seabed geology.

Assessing Local Low-level Features with Segmentation

Classification of images may help to facilitate object recognition in robotic vision. Computers process pixel information in images for image representation.There are various forms of pixel information which are called low-level features that can be extracted from images. Different feature schemes may suit well in different types of images. Hence, identifying features for image representation is an important step. This paper presents a work in accessing local low-level features to facilitate image classification particularly in natural wildlife images. The images first go through segmentation and every segment is then tiled into image blocks. Different types of image features are extracted from the image segments and also image blocks. The assessment on the local low-level features in image classification is conducted to identify the best image features for image representation in some sample image sets. It was found that the best classification comes from the image blocks represented by the concatenation of LUV colour and Haralick texture features in our dataset.

Image microarrays derived from tissue microarrays (IMA-TMA): New resource for computer-aided diagnostic algorithm development

Background: Conventional tissue microarrays (TMAs) consist of cores of tissue inserted into a recipient paraffin block such that a tissue section on a single glass slide can contain numerous patient samples in a spatially structured pattern. Scanning TMAs into digital slides for subsequent analysis by computer-aided diagnostic (CAD) algorithms offers the possibility of evaluating candidate algorithms against a near-complete repertoire of variable disease morphologies. This parallel interrogation approach simplifies the evaluation, validation, and comparison of such candidate algorithms. A recently developed pair of digital tools, digital core (dCORE),and image microarray maker (iMAM) enables the capture of uniformly sized and resolution-matched images, with these representing key morphologic features and fields of view, aggregated into a single monolithic digital image file in an array format, which we define as an image microarray (IMA). We further define the TMA-IMA construct as IMA-based images derived from whole slide images of TMAs themselves. Methods: Here we describe the first combined use of the previously described dCORE and iMAM tools, towards the goal of generating a higher-order image construct, with multiple TMA cores from multiple distinct conventional TMAs assembled as a single digital image montage. This image construct served as the basis of the carrying out of a massively parallel image analysis exercise, based on the use of the previously described spatially invariant vector quantization (SIVQ) algorithm. Results: Multicase, multifield TMA-IMAs of follicular lymphoma and follicular hyperplasia were separately rendered, using the aforementioned tools. Each of these two IMAs contained a distinct spectrum of morphologic heterogeneity with respect to both tingible body macrophage (TBM) appearance and apoptotic body morphology. SIVQ-based pattern matching, with ring vectors selected to screen for either tingible body macrophages or apoptotic bodies, was subsequently carried out on the differing TMA-IMAs, with attainment of excellent discriminant classification between the two diagnostic classes. Conclusion: The TMA-IMA construct enables and accelerates high-throughput multicase, multifield based image feature discovery and classification, thus simplifying the development, validation, and comparison of CAD algorithms in settings where the heterogeneity of diagnostic feature morphologic is a significant factor.

Significance of Textural Features in Aerial Images

Texture is one of the important characteristics used in identifying objects or regions of interest in an image, whether the image is a photomicrograph, an aerial photograph, or a satellite image. However, in order to identify and characterize tree species, the resolution of satellite imagery is still not high enough. It is necessary to acquire higher resolution data using e.g. an airplane. Image features for classification of tree species were extracted by texture analysis using Gray Level Co-occurrence Matrix (GLCM). In this analysis, Textural features were computed from the GLCM of image. This textural feature can be used for classifying tree species in aerial image. At last, we compare the textural features performance on two different sequences of same aerial image. We analyze all the simulations have been performed on MATLAB 7.4.0.287 (R2007a).

Improved PCA + LDA Applies to Gastric Cancer Image Classification Process

Abstract Principal component analysis (PCA) and linear discriminant analysis (LDA) are two most widely used pattern recognition methods in the field of feature extraction,while PCA + LDA is often used in image recognition.Here,we apply PCA + LDA to gastric cancer image feature classification, but the traditional PCA + LDA dimension reduction method has good effect on the training sample dimensionality and clustering, the effect on test samples dimension reduction and clustering is very poor, that is, the traditional PCA + LDA exists Generalization problem on the test samples. To solve this problem, this paper proposes an improved PCA + LDA method, which mainly considers from the LDA transform; improves the traditional PCA + LDA;increase the generalization performance of LDA on test samples and increases the classification accuracy on test samples. The experiment proves that the method can achieve good clustering.

Carotid artery recognition system: A comparison of three automated paradigms for ultrasound images

The development of completely automated techniques for arterial wall segmentation and intima-media thickness measurement requires the recognition of the artery in the image frame. Conceptually, automated techniques can be thought of as the combination of two cascaded stages: artery recognition and wall segmentation. In this paper, the authors show three carotid artery recognition systems (CARS) that are fully automated. The first technique is based on a first-order derivative Gaussian edge analysis (CARSgd). The second method is based on an integrated approach (CARSia) that combines image feature extraction, fitting, and classification. The third strategy is based on signal analysis (CARSsa). The output of all the three paradigms provide tracing of the far adventitial (AD(F)). The authors validated CARSgd, CARSia, and CARSsa on a dataset of 365 longitudinal B-Mode carotid images, acquired by different sonographers. Performance evaluation of the carotid recognition process was done in three ways: (1) visual inspection by experts; (2) by measuring the Hausdorff distance (HD) between the automatic far adventitial (AD(F)) and the manually traced AD(F), and (3) by measuring the HD between AD(F) and the lumen-intima (GT(LI)) and media-adventitia (GT(MA)) borders of the arterial walls. The average HD between AD(F) and the manual AD(F) was 1.53 ± 1.51 mm for CARSgd, 1.82 ± 3.08 mm for CARSia, and 2.56 ± 2.89 mm for CARSsa. The average HD between GT(LI) and AD(F) for CARSgd, CARSia, and CARSsa were 2.16 ± 1.16 mm, 2.71 ± 2.89 mm, and 2.66 ± 1.52 mm, respectively. The average HD between AD(F) and GT(MA) for CARSgd, CARSia, and CARSsa were 1.54 ± 1.19 mm, 1.86 ± 2.66 mm, and 1.95 ± 1.64 mm, respectively. Considering a maximum distance of 50 pixels (about 3 mm), CARSgd showed an identification accuracy of 100%, CARSia of 92%, and CARSsa of 96%. These identification accuracies were confirmed by visual inspection. All the three systems work on MATLAB, Windows OS, and on a PC based cross platform medical application written in Java called ATHEROEDGE™ with 1 s per image. CARSgd showed very accurate AD(F) profiles coupled with a low computational burden and without the need for specific tuning. It can be thought of as a reference technique for carotid localization, to be used in automated intima-media thickness measurement strategies.

Monitoring Urban Tree Cover Using Object-Based Image Analysis and Public Domain Remotely Sensed Data

Urban forest ecosystems provide a range of social and ecological services, but due to the heterogeneity of these canopies their spatial extent is difficult to quantify and monitor. Traditional per-pixel classification methods have been used to map urban canopies, however, such techniques are not generally appropriate for assessing these highly variable landscapes. Landsat imagery has historically been used for per-pixel driven land use/land cover (LULC) classifications, but the spatial resolution limits our ability to map small urban features. In such cases, hyperspatial resolution imagery such as aerial or satellite imagery with a resolution of 1 meter or below is preferred. Object-based image analysis (OBIA) allows for use of additional variables such as texture, shape, context, and other cognitive information provided by the image analyst to segment and classify image features, and thus, improve classifications. As part of this research we created LULC classifications for a pilot study area in Seattle, WA, USA, using OBIA techniques and freely available public aerial photography. We analyzed the differences in accuracies which can be achieved with OBIA using multispectral and true-color imagery. We also compared our results to a satellite based OBIA LULC and discussed the implications of per-pixel driven vs. OBIA-driven field sampling campaigns. We demonstrated that the OBIA approach can generate good and repeatable LULC classifications suitable for tree cover assessment in urban areas. Another important finding is that spectral content appeared to be more important than spatial detail of hyperspatial data when it comes to an OBIA-driven LULC.

Measuring the relative extent of pulmonary infiltrates by hierarchical classification of patient-specific image features

Pulmonary infiltrates are common radiological findings indicating the filling of airspaces with fluid, inflammatory exudates, or cells. They are most common in cases of pneumonia, acute respiratory syndrome, atelectasis, pulmonary oedema and haemorrhage, whereas their extent is usually correlated with the extent or the severity of the underlying disease. In this paper we propose a novel pattern recognition framework for the measurement of the extent of pulmonary infiltrates in routine chest radiographs. The proposed framework follows a hierarchical approach to the assessment of image content. It includes the following: (a) sampling of the lung fields; (b) extraction of patient-specific grey-level histogram signatures from each sample; (c) classification of the extracted signatures into classes representing normal lung parenchyma and pulmonary infiltrates; (d) the samples for which the probability of belonging to one of the two classes does not reach an acceptable level are rejected and classified according to their textural content; (e) merging of the classification results of the two classification stages. The proposed framework has been evaluated on real radiographic images with pulmonary infiltrates caused by bacterial infections. The results show that accurate measurements of the infiltration areas can be obtained with respect to each lung field area. The average measurement error rate on the considered dataset reached 9.7% ± 1.0%.

Historical Land Use as a Feature for Image Classification

This paper analyzes the effect of the addition of historical land-use as a descriptive feature in plot-based image classification when updating land-use/land-cover geospatial databases. Several historical databases have been simulated to assess the influence and significance of this feature in the classification. The causes, nature, and evolution of classification errors as the database currency varies are analyzed; and the impact of these errors on change detection during the updating process is evaluated. The results show that the addition of historical land-use information increases the overall accuracy of image classifications. During a database updating process, changes are detected by comparing the historical land-use with the classification results. The main drawback of employing historical land-use as a descriptive feature in image classification for change detection is that the percentage of undetectable errors significantly increases as more accurate is the database information.

Image modeling with combined optimization techniques for image semantic annotation

Image semantic annotation can be viewed as a multi-class classification problem, which maps image features to semantic class labels, through the procedures of image modeling and image semantic mapping. Bayesian classifier is usually adopted for image semantic annotation which classifies image features into class labels. In order to improve the accuracy and efficiency of classifier in image annotation, we propose a combined optimization method which incorporates affinity propagation algorithm, optimizing training data algorithm, and modeling prior distribution with Gaussian mixture model to build Bayesian classifier. The experiment results illustrate that the classifier performance is improved for image semantic annotation with proposed method.