Sparse representation with multi-manifold analysis for texture classification from few training images

Texture classification is a classical yet still active topic in computer vision and pattern recognition. Recently, several new texture classification approaches by modeling texture images as distributions over a set of textons have been proposed. These textons are learned as the cluster centers in the image patch feature space using the K-means clustering algorithm. However, the Euclidian distance based the K-means clustering process may not be able to well characterize the intrinsic feature space of texture textons, which if often embedded into a low dimensional manifold. Inspired by the great success of l <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sub> -norm minimization based sparse representation (SR), in this paper we propose a novel texture classification method via patch-based sparse texton learning. Specifically, the dictionary of textons is learned by applying SR to image patches in the training dataset. The SR coefficients of the test images over the dictionary are used to construct the histograms for texture classification. Experimental results on benchmark database validate the effectiveness of the proposed method.

Using statistical textons for texture classification has shown great success recently. The maximal response 8 (Statistical_MR8), image patch (Statistical_Joint) and locally invariant fractal (Statistical_Fractal) are typical statistical texton algorithms and state-of-the-art texture classification methods. However, there are two limitations when using these methods. First, it needs a training stage to build a texton library, thus the recognition accuracy will be highly depended on the training samples; second, during feature extraction, local feature is assigned to a texton by searching for the nearest texton in the whole library, which is time consuming when the library size is big and the dimension of feature is high. To address the above two issues, in this paper, three binary texton counterpart methods were proposed, Binary_MR8, Binary_Joint, and Binary_Fractal. These methods do not require any training step but encode local feature into binary representation directly. The experimental results on the CUReT, UIUC and KTH-TIPS databases show that binary texton could get sound results with fast feature extraction, especially when the image size is not big and the quality of image is not poor.

The method of texture feature extraction and classification of pork loin B ultrasound image is proposed, which can be applied to the computer-aided judgment the pork loin fat content of pork loin. 5 texture features which is based on the texture of the co-occurrence matrix are extracted from the B ultrasound image of pork loin according to the digital image processing algorithm. Using the correlation analysis method to select the key texture extraction in the first step. Then,the classification is realized based on the BP neural network. The train set and test set are randomly chosen from 135 cases. Tests performed show that the proposed method result in a high classification accuracy, which will provide the researcher a valuable opinion on the pork fat content detection.

In information processing using Wavelet transform, wavelet subband coefficients are often modelled by a probability distribution function. Recently, a local energy histogram method has been proposed to alleviate the difficulty in modeling wavelet subband coefficients with a previously assumed distribution function. Actually, the similarity between any two local energy histograms was measured by a symmetrized Kullback-Leibler divergence (SKLD). However, this measurement neglects the balance of wavelet subbands’ roles in texture classification. In this paper, we propose an efficient texture classification method based on weighted symmetrized Kullback-Leibler divergences (WSKLDs) between two local energy histograms (LEHs). In particular, for any test and training images, we index their Wavelet subbands in the same way, and weight the SKLD between any two LEHs of the s-th wavelet subbands of two image by the reciprocal of the summation of the SKLDs between the expected LEHs of any two different texture classes over all training images. Experimental results reveal that our proposed method outperforms five state-of-the-art methods.KeywordsTexture classificationWavelet subbandImbalance problemLocal energy histogram (LEH)Kullback-Leibler divergence (KLD)

Aiming at the problems of traditional sparse representation algorithms that has a large amount of calculation and low recognition rate under the condition of small samples, a sparse representation classification algorithm (PCA-SRC) combined with principal component feature extraction is proposed. The algorithm can maintain the manifold structure of the finger vein space, solve the singularity problem of the characteristic equation matrix in small sample image recognition. Principal component analysis (PCA) is used to reduce the dimension of the feature matrix to eliminate redundancy. Construct an over-complete dictionary of sparse representation, and design a classifier based on sparse representation to get the classification results. The finger vein feature matching experiment is carried out using the self-built finger vein data set, and the algorithm proposed in this paper is compared with the traditional algorithm. The results show that the recognition rate of the four algorithms can reach more than 94%. The fusion of PCA and SRC speeds up the recognition speed. The recognition rate is higher than the traditional algorithm under different numbers of training samples, and the best recognition rate is 100%. The algorithm has high finger vein recognition rate and low dependence on the training sample set, which provides a new idea for the realization of finger vein recognition.

Spectral Filter Array cameras provide a fast and portable solution for spectral imaging. Texture classification from images captured with such a camera usually happens after a demosaicing process, which makes the classification performance rely on the quality of the demosaicing. This work investigates texture classification methods applied directly to the raw image. We trained a Convolutional Neural Network and compared its classification performance to the Local Binary Pattern method. The experiment is based on real SFA images of the objects of the HyTexiLa database and not on simulated data as are often used. We also investigate the role of integration time and illumination on the performance of the classification methods. The Convolutional Neural Network outperforms other texture classification methods even with a small amount of training data. Additionally, we demonstrated the model’s ability to adapt and scale for different environmental conditions such as illumination and exposure compared to other methods. In order to explain these results, we analyze the extracted features of our method and show the ability of the model to recognize different shapes, patterns, and marks in different textures.

Deep neural networks have greatly improved the performance of rail surface defect segmentation when the test samples have the same distribution as the training samples. However, in practical inspection scenarios, the rail surface exhibits variations in appearance due to different service times and natural conditions. Conventional deep learning models show limited generalization in scenes with distribution differences. To address this problem, we propose a novel one-shot unsupervised domain adaptation framework. Specifically, we introduce a shape-consistent style transfer module that performs pixel-level distribution alignment between the training and test images. Based on the one-shot test image, the training image is reconstructed to have the same appearance as the test image. Meanwhile, we employ a multitask learning strategy to prevent content distortion of the reconstructed images. To improve the robustness of the model to distribution differences, we design an edge-aware defect segmentation model and train the model using the reconstructed training images. The experimental results show that our method effectively improves the robustness of the model to distribution differences and achieves satisfying results in the task of rail surface defect segmentation.

Trial-by-trial texture classification analysis and identifying salient texture related EEG features during active touch that are minimally influenced by movement type and frequency conditions are the main contributions of this work. A total of twelve healthy subjects were recruited. Each subject was instructed to use the fingertip of their dominant hand’s index finger to rub or tap three textured surfaces (smooth flat, medium rough, and rough) with three levels of movement frequency (approximately 2, 1 and 0.5 Hz). EEG and force data were collected synchronously during each touch condition. A systematic feature selection process was performed to select temporal and spectral EEG features that contribute to texture classification but have low contribution towards movement type and frequency classification. A tenfold cross validation was used to train two 3-class (each for texture and movement frequency classification) and a 2-class (movement type) Support Vector Machine classifiers. Our results showed that the total power in the mu (8–15 Hz) and beta (16–30 Hz) frequency bands showed high accuracy in discriminating among textures with different levels of roughness (average accuracy > 84%) but lower contribution towards movement type (average accuracy < 65%) and frequency (average accuracy < 58%) classification.

Multiresolution histograms have been recently proposed as robust and efficient features for texture classification. In this paper, we evaluate the performance of multiresolution histograms for texture classification using support vector machines (SVMs). We observe that the dimensionality of multiresolution histograms can be greatly reduced with a Laplacian pyramidal decomposition. With an appropriate kernel, we show that SVMs significantly improve the performance of multiresolution histograms compared to the previously used nearest-neighbor (NN) classifiers on a texture classification problem involving Brodatz textures. Experimental results indicate that multiresolution histograms in conjunction with SVMs are also robust to noise.

Principal curvatures based rotation invariant algorithms for efficient texture classification

ICA (Independent Component Analysis) is a mathematical tool traditionally employed for source separation. In this paper, we test its ability for texture analysis, in order to provide a new texture classification method. From the multitude of the existing algorithms, we have chosen FastICA, a version based on the forth order statistics of the analyzed signal. By FastICA, a texture is decomposed in a weighted sum of components with a rather high degree of independence. Each component is further described by means of its negentropy, which is a measure of the nongaussianity. We show experimentally, that the three most nongaussian components of each analyzed texture are able to cluster the test samples.

Texture classification algorithms were employed widely in analyzing remote sensing images, medical images, and industrial images testing and so on. There are many texture classification methods have been developed which including K-Vive based algorithms. Among of them, K-view-voting method achieved best performance by comparing with other K-view based methods. In this paper, after analyzing K-view-voting method, an improved method, which performed in the process of decision, was proposed to improve the classification performance of K-view-voting. To test and evaluate the performance of new method compared with K-view-voting method, several experiments were conducted. The testing images in dataset were selected randomly in a standard database. Experimental results demonstrated the effect of the new method.

In this paper, we propose a texture classification method using local texture features BDIP (block difference of inverse probabilities) and BVLC (block variation of local correlation coefficients) in wavelet domain. BDIP and BVLC are known to be good texture features which are bounded and well normalized to reduce the effect of illumination and catch the own properties of textures effectively. In the method, a target image is first decomposed into wavelet subbands. BDIPs and BVLCs are then computed in wavelet subbands. The means and standard deviations of subband BDIPs and BVLCs and the subband standard deviations are fused into a texture feature vector. Finally, the Bayesian distance between the feature vector of a query image and that of each class is stably measured and it is classified into the class of minimum distance. Experimental results for three test databases (DBs) show the proposed method yields excellent performances.

This paper presents a new method for texture image classification based on compressed sensing (CS) and feature representation using scale-invariant Bag-of-Words model (SIBW). The issue is discussed from the following three aspects: 1) texture image compression, 2) Scale Invariant Feature Transform (SIFT) feature extraction in compressed domain and 3) classification by using SIBW. The texture classification method was tested on six common classes of texture images, the results show that the excellent performance can be achieved by the proposed approach, and classification accuracy has been greatly improved.

Sparse representation-based single image super-resolution (SISR) methods use a coupled overcomplete dictionary trained from high-resolution images/image patches. Since remote sensing (RS) satellites capture images of large areas, these images usually have poor spatial resolution and obtaining an effective dictionary as such would be very challenging. Moreover, traditional patch-based sparse representation models for reconstruction tend to give unstable sparse solution and produce visual artefact in the recovered images. To mitigate these problems, in this article, we have proposed an adaptive joint sparse representation-based SISR method that is dependent only on the input low-resolution image for dictionary training and sparse reconstruction. The new model combines patch-based local sparsity and group sparse representation-based nonlocal sparsity in a single framework, which helps in stabilizing the sparse solution and improve the SISR results. The experimental results are evaluated both visually and quantitatively for several RGB and multispectral RS datasets, where the proposed method shows improvements in peak signal-to-noise ratio by 1–4 dB and 2–3 dB over the state-of-the-art sparse representation- and deep learning-based SR methods, respectively. Land cover classification applied on the super-resolved images further validate the advantages of the proposed method. Finally, for practical RS applications, we have performed parallel implementation in general purpose graphics processing units and achieved significant speed ups (30–40×) in the execution time.