Texture Feature Extraction

ABSTRACT: Image content based features analysis is the emerging area of research in medical imaging and in large databases. Texture is one of the important characteristics used for the analysis of medical images and interest regions in the image. With the development in the technology and image processing algorithms like wavelet based multi- resolution analysis and Gabor filter methods provided the platform to extract the texture features in an effective way. This paper aims at describing different approaches currently used for image based retrieval and feature extraction.Texture feature extraction maps the spatial structural patterns either stochastic or geometric in differences of gray level in the image. Different texture measures are presented and analyzed with categorization. Later wavelet based method has been chosen and the statistical features has been extracted with parameters such as mean and standard deviation.

Dermatology is the part of the drug that manages skin, hair, and nails in the amplest sense. To propose a framework which would help the patients just as specialists to analyze the skin illnesses by simply giving the picture of the influenced region of skin. proposed location framework utilizes texture feature extraction strategies GLCM and GLRLM alongside ANN strategy for classifying the skin diseases. GLCM extract the features are contrast, correlation, homogeneity, and Energy. GLRLM extract the features are SRE, LRE, RP, GLNU, RLNU, LGRE, HGRE. The Skin diseases classification system comprises of three sections where the initial section is to train the folder of images. In training, surface highlights are chosen from the training images set by utilizing GLCM and GLRLM techniques after the change of RGB images into grayscale and store the pictures into a (.mat) table. In the second part is testing the images. The testing stage goes about as approval. In the testing stage, accepts one picture as information at that point convert the RGB picture into grayscale. GLCM and GLRLM texture feature extraction strategies are utilized. After feature extractions, store the incentive in another (.mat) table. The third part is the classification part. ANN classification used to classify skin diseases such as Eczema, Melanoma, Dermatitis, Basal cell carcinoma, and Acne. Compare the accuracy levels both GLCM and GLRLM methods.

Content based image retrieval (CBIR) is a challenging problem due to large size of the image database, difficulty in recognizing images, difficulty in devising a query and evaluating results in terms of semantic gap, computational load to manage large data files and overall retrieval time. Feature extraction is initial and important step in the design of content based image retrieval system. Feature extraction is a means of extracting unique and valuable information from the image. These features are also termed as signature of image. Feature extraction of the image in the database is done offline therefore it does not contribute significantly in computational complexity. Humans tend to differentiate images based on color, therefore color features are mostly used in CBIR. Color moment is mostly used to represent color features especially when image contain just an object. Regularity, directionality, smoothness and coarseness are some of the texture properties perceived by human eye. Gabor filter and wavelet transform for texture feature extraction has proved to be very effective in describing visual content via multi-resolution analysis. The paper mainly gives the brief ideas of existing retrieval techniques. Also paper gives the comparative analysis of mentioned techniques with different metrics.

The conventional method of leaf classification involves two main steps. The first step is obtaining a priori knowledge of each class to be recognized. Normally this knowledge encompasses some sets of texture feature of one or all of the classes. Once the knowledge is available and texture feature of the observed image are extracted, then classification techniques, for example nearest neighbors and decision trees, can be used to make the decision that is the second step. Such a procedure is illustrated the tasks that texture classification has been applied to include the classification of plant leaves images. Currently there are a huge number of texture feature extraction methods available and most of the methods are associated with tunable parameters. It is difficult to find the most suitable. Patten recognition, the k-nearest neighbors Algorithm (k-NN) is a non-parametric method used for classification and regression.

In this paper texture feature extraction for multispectral satellite image using GLCM has been used to identify the change detection. In preprocessing the image was converted into hex file because the hardware module does not understand the image. GLCM calculation unit used to extract the gray level co-occurrence matrix from the image. Four second order features namely angular second moment, correlation, energy, contrast are calculated from the GLCM matrix. Finally compare the feature parameters of hardware and software results. This process is explained in detailed manner as follows.

It is easily noticeable that signal processing methods are very popularly used in the recent years, especially for Gabor filters and wavelets. Although these methods require more computation as they are examining the frequency domain, the accuracy obtained is good and usually outperform older and simpler techniques. The old technique like GLCM is however yet to be forgotten in the field of texture classification because it is one of the simplest textural feature which is old but is computationally inexpensive. It remains to be mainly used as a baseline algorithm for comparative studies especially when a new application of texture classification is experimented. The GLCM is however more commonly used in some improved or combined ways recently but none of these variants have grown into a major trend.

Segmentation as the name suggest refers to the process of partitioning a digital image into many segments (which is sets of pixels a lso known as super pixels). The aim of segmentation is to simplify and/or change the representation of image into something that is more meaningful and easier to analyze. Image segmentation which is typically used to locate objects and boundaries (lines, curves, etc.) in images. More precisely, image segmentation is the process of assigning a label to every pixel in an image such that pixels with the same label share certain visual characteristics. The image segmentation results into a set of segments that collectively cover the whole image, or a set of contours extracted from the image. Each pixel in a region is similar with respect to some characteristic or computed property, such as color, intensity, or texture. Adjacent regions are significantly different with respect to the same characteristic.

A major component in analyzing images involves data reduction which is accomplished by intelligently modifying the image from the lowest level of pixel data into higher level representations. From these higher level representations we can gather useful information; a process called feature extraction [9]. Extracting features by fixed blocs in the image has been considered to be sufficient as an ROI selection method in some medical applications where a large fraction of the image is covered by tissue of interest.

Although the SGLDM is one of the best general texture analysis methods and has been widely used for texture characterisation, the employment of the co­ occurrence matrices for storing the estimated second-order probabilities has a lot of disadvantages. These disadvantages limit its applicability and prevent the ex­ traction of all the texture information from a given image th a t can be captured by the SGLDM. A lot of authors have referred to those disadvantages which are the memory requirements of the co-occurrence m atrix approach and the compu­ tational time for calculating textural features from these matrices. Many of the researchers who used this method, were forced to reduce the grey level range of the image, so th at they could handle the large requirements of the co-occurrence matrices. Doing so, however, they lost significant texture information in many cases, which could have led to better texture discrimination. Others employed a limited set of texture features to reduce the computational time complexity. Still others completely abandoned this method due to these requirements and shifted towards other texture analysis methods which, though, were shown not to be equally powerful.

Abstract: To detect the correct tumor from an MRI image is a difficult task as it is done manually. So automatic detection of brain tumor plays very important role in medical as well as in image processing. While identifying tumor from MRI images that image will go through the number of process and feature extraction is one of those method which help to increase the accuracy of detection and segmentation. Multifractal is one of the features which describe the self similarities preset in an image. This fractal geometry describes the irregular or discontinuous shape of natural features and complex objects that traditional Euclidean geometry is unable to analyze. The combination of Multifractal dimension with intensity feature will help to improve the detection and segmentation of brain tumor .

Textures are a pattern of non-uniform spatial distribution of differing image intensities [11], which focus mainly on the individual pixels that make up an image. Texture is defined by quantifying the spatial relationship between materials in an image. Texture analyses is done by using various approaches like statistical, fractal, and structural. Statistical type includes techniques like grey-level histogram, grey-level co-occurence matrix [20], local binary pattern operator, auto-correlation features and power spectrum.

where I(s) represents the unit measure, in this case the gray-level intensity, s the scale used, and D is the fractal dimension (Hausdorff-Besicovitch dimension). In texture representation, FD alone does not fully represent a rich description. Different textures may have the same FD due to combined differences in directionality and coarseness [20]. These uncertainties can be addressed by multifractal analysis [21, 15], where a point categorization is defined on the object function based on some criteria. The FD is estimated for every point set according to this categorization. A common criteria for categorization is the probably density function estimated from the image intensity [15, 16].

A. Majumdar et al (2009)[12] done a comparative study of curvelets , wavelets and contourlets. Curvelets, wavelets and contourlets used as a feature sets. Mostly used in pattern recognition. In this curvelets and contourlets are multi- resolution multidirectional transforms, they are used in face and character recognition [12]. Comparison result obtained is, for higher resolutions the wavelets used is a good option as a feature set for facial as well as character image. But as the resolution starts decreasing wavelets not work efficiently as feature descriptors. But at lower resolution curvelets work very efficiently. Contourlets not useful in higher as well as lower resolution, it only works in very coarse resolution. Recognition capabilities of all these three are tested using KNN classification [12].

This work presents a simple process for gray image colorization, using a colored image which is similar to this gray scale image but not the colorized version of the gray image. This colored image is retrieved from the data base of colored images that has been created for this purpose. Here, the texture properties of the colored images are extracted and stored. For the purpose of colorization these features are compared with those of the gray image to be colorized and the best matching image is found out from the database. For colorization of this gray scale image a decorrelated color space YCbCr is utilized. This technique is completely automatic and no human intervention is required in the process of colorization. Apart from this the technique presented here is very fast and produces good quality results as compared to the conventional colorization methods. Texture features used here to calculate a texture similarity measure are energy, entropy, contrast, homogeneity, autocorrelation based on correlation matrix as well as coarseness and directionality.

In computer graphics, texture synthesis is a common technique to create large textures from usually small texture samples, for the use of texture mapping in surface or scene rendering applications. A synthetic texture should differ from the samples, but should have perceptually identical texture characteristics [DeB97]. The main advantage of texture synthesis in this case is that it can naturally handle boundary condition and avoid verbatim repetitions. In computer vision, texture synthesis is of interest also because it provides an empirical way to test texture analysis. Because a synthesis algorithm is usually based on texture analysis, the result justifies effectiveness of the underlying models. Compared to texture classification and segmentation, texture synthesis poses a bigger challenge on texture a nalysis because it requires a more detailed texture description and also reproducing textures is generally more difficult than discriminating them.

To sum up the literature review, we can see there is a lack of references devoted to skin image retrieval by computer extracted visual features for various types of skin disease. Skin image retrieval differentiates from disease justification between only two possibilities like melanoma research, or image classification – classification of a group of images into different categories, or retrieval by text. However, we can still borrow ideas from those techniques, such as the use of color, texture feature extraction, and

In this paper, an improved machine learning algorithm is proposed for jaundice detection. A common condition in newborns, jaundice refers to the yellow color of the skin and whites of the eyes caused by excess bilirubin in the blood. In this research, preprocessing is done by using the hybrid median filtering technique and GLCM is used for texture feature extraction and color feature extraction. In this analysis, an ensemble of fitness evaluations would produce an ensemble of fitness values for each individual. Feature ranking is useful to gain knowledge of data and identify relevant features. The ranking of individual features will be done, in order to reduce the GA optimization complexity. After ranking of features, the top most features will be inserted into GA. Finally, the Kernel Support Vector Machine (SVM) is used to classify the normal babies and jaundice detected babies.

applied the SIFT features on binary images and keypoints from the images are used in k-means clustering to reduce the feature dimensions. Jerrin Varghese (2015) studied image search based on scale invariant feature transform descriptors using k-means clustering algorithm. N. Puviarasan, et al. (2014) proposes a combined shape and texture feature extraction technique for content based image retrieval system. Aiysha Begam, et al. (2013) proposes a CBIR system exploitation the local color and texture options of chosen image sub-blocks and world color and form options of the image. A combined color and texture feature is computed for every region. Hiran Ganegedara, et al. (2012) proposed Parallel GSOM algorithm has demonstrated that parallel computation can significantly reduce training time for self- organizing maps. Hsin-Chien Huang, et al. (2012) proposed an affinity aggregation spectral clustering algorithm ring, SIFT, spectral clustering. For aggregating affinity matrices for spectral clustering, it was more immune to ineffective affinities and irrelevant features. Also, it enables the construction of similarity measures for clustering less crucial. Nenad Tomašev, et al. (2011) explores the ways to represent images as bags of SIFT feature clusters and created a hybrid clustering algorithm which offers more flexibility than simple spatial k-means clustering. N.Nanthini, et al. (2017) proposed a combination of histogram and SIFT feature extraction technique with spectral clustering algorithm to retrieve images similar to query image. The performance of the proposed feature extraction technique is higher than the separate feature.

Abstract— Texture analysis plays an increasingly important role in computer vision. Since the textural properties of images appear to carry useful information for discrimination purposes, it is important to develop significant features for texture. Various texture feature extraction methods include those based on gray-level values, transforms, auto correlation etc. We have chosen the Gray Level Co occurrence Matrix (GLCM) method for extraction of feature values. Image segmentation is another important problem and occurs frequently in many image processing applications. Although, a number of algorithms exist for this purpose, methods that use the Expectation-Maximization (EM) algorithm are gaining a growing interest. The main feature of this algorithm is that it is capable of estimating the parameters of mixture distribution. This paper presents a novel unsupervised segmentation method based on EM algorithm in which the analysis is applied on vector data rather than the gray level value.

system uses digital image processing techniques to identify normal and abnormal chest x-rays. Motivation behind this project is mass screening of large population that is not feasible manually. K-means clustering segmentation is used to segment the lung part from chest radiograph. Shape and texture feature extraction method is used to extract shape and texture features from segmented lung field and Euclidean distance classification is used to classify chest x- ray into normal or abnormal x-ray. System shows 70% accuracy on test image dataset and 70% specificity whereas sensitivity of system is 70% on test image dataset. The paper is divided in to 6 sections. Section 2 describe the literature review of the implemented system. Section 3 presents the information about image dataset used in the experiments. In Section 4, detailed discussion of lung segmentation, feature extraction and classification is presented. A results of practical experiments follows in Section 5. Finally, conclusion concludes the paper.