A Novel CBIR-Based System using Texture Fused LBP Variants and GLCM Features
Meenakshi Garg1, Manisha Malhotra2, Harpal Singh3
1Meenakshi Garg*, University Institute of Computing, Chandigarh University, Mohali, Punjab, India.
2Manisha Malhotra, University Institute of Computing, Chandigarh University, Mohali, Punjab, India.
3Harpal Singh, Department of Electronics and Communication Engineering, CEC Landran, Punjab, India.
Manuscript received on November 14, 2019. | Revised Manuscript received on 23 November, 2019. | Manuscript published on December 10, 2019. | PP: 1243-1247 | Volume-9 Issue-2, December 2019. | Retrieval Number: B6373129219/2019©BEIESP | DOI: 10.35940/ijitee.B6373.129219
Open Access | Ethics and Policies | Cite | Mendeley | Indexing and Abstracting
© The Authors. Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open access article under the CC-BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Abstract: This paper presents Multiple-features extraction and reduction for Content-Based Image Retrieval (CBIR). At first, discrete wavelet transform (DWT) is applied on the R, G and B channels to get multi-level decomposition of the image in which approximation and detailed coefficients are extracted. Approximation coefficients contain the main content of the image while detailed coefficients provide the local noise variation in the image. Further computationally efficient and rotationinvariant Dominant-Rotated Local Binary Pattern called a texture descriptor is applied on all approximation and detailed coefficients. By calculating the descriptor relative to reference in a local neighbor patch, a rotation invariance feature image is obtained. The proposed methodology contains the complete structural information extracted by the Local Binary Patterns and also extract the extra information using the information of magnitude, thus attaining extra discriminative power. Then, by getting the Dominant Rotated Local Binary Pattern image, concept of GLCM has been used to extract the statistical characteristics for the classification of texture images. GLCM directly works with the intensity of the images and also provides the spatial relationship of the pixels in the image by calculating frequency of occurring of similar patterns in different directions which makes it useful for the extraction of texture characteristics. GLCM has been improvised into a generalized co-occurrence matrix that extracts significant spatial properties from the distribution of local maxima. Further Median Robust Extended Local Binary Pattern is extracted out of the approximation and detailed coefficients and also the histogram is calculated out of them. Unlike the traditional LBP method and several variants of the LBP, Median Robust Extended Local Binary Pattern compares the local median of images instead of intensities of the raw images. It is a multiscale LBP-type descriptor that proficiently compares the image medians with novel sampling schemes, capable of capturing both microstructure and macrostructure. Further feature concatenation is applied in which GLCM features of DRLBP and histogram of MRELBP are combined for getting a large feature vector. Further, we applied a mutual information concept to sort out the most differentiable features for all categories of CORAL dataset which are further fed to particle swarm optimizationbased feature selector which reduced the number of features that can be used in the classification phase. PSO uses the SVM classifier in evaluating its objective function which is average precision value for the selected features. PSO tries to increase precision value for all the categories and provides a feature vector with a large precision value when classify by SVM. Further three classifiers are trained ad tested named SVM, KNN and decision tree in which SVM gives high accuracy and precision rates of classification. Experimental results show above 94% accuracy and .80 to .90 precision values for most of the categories of CORAL dataset.
Keywords: CBIR, MLBP, DRLBP, PSO, SVM, Feature Selection, Classification etc.
Scope of the Article: Classification