A Novel Discriminating and Relative Global Spatial Image Representation with Applications in CBIR

Abstract

The requirement for effective image search, which motivates the use of Content-Based Image Retrieval (CBIR) and the search of similar multimedia contents on the basis of user query, remains an open research problem for computer vision applications. The application domains for Bag of Visual Words (BoVW) based image representations are object recognition, image classification and content-based image analysis. Interest point detectors are quantized in the feature space and the final histogram or image signature do not retain any detail about co-occurrences of features in the 2D image space. This spatial information is crucial, as it adversely affects the performance of an image classification-based model. The most notable contribution in this context is Spatial Pyramid Matching (SPM), which captures the absolute spatial distribution of visual words. However, SPM is sensitive to image transformations such as rotation, flipping and translation. When images are not well-aligned, SPM may lose its discriminative power. This paper introduces a novel approach to encoding the relative spatial information for histogram-based representation of the BoVW model. This is established by computing the global geometric relationship between pairs of identical visual words with respect to the centroid of an image. The proposed research is evaluated by using five different datasets. Comprehensive experiments demonstrate the robustness of the proposed image representation as compared to the state-of-the-art methods in terms of precision and recall values.

An efficient content based image retrieval using enhanced multi-trend structure descriptor

“…Concurrently, Gradient location and orientation histogram with PCA is used in Mikolajczyk and Schmid [39] for retrieving the images; Liu and Yang [40] computes the spatial correlation of textons using texton co-occurrence matrices (TCM) which extracts energy, entropy, contrast and homogeneity to represent the image; attributes of cooccurrence matrix is expressed using histogram based on Julesz's textons theory for analyzing the natural images in Liu et al [41] and is named as multi-texton histogram (MTH), and the authors confirmed that their approach achieves better performance than the texton co-occurrence matrix and edge orientation auto-correlogram; feature based on edge orientation similarity and underlying colors is described by Liu et al [42] and named it as Micro-Structure Descriptor (MSD) which captures local level color and texture effectively; Saliency Structure Histogram (SSH) reported in Liu and Yang [43] computes the logarithm characteristics of Gabor energy to describe the image; Structure Element Descriptor (SED) comprising of color and texture information and Structure Element Histogram (SEH) comprising of the spatial correlation of color and texture feature is reported in Xingyuan and Zongyu [44] for image retrieval; Seetharaman and Sathiamoorthy [45] introduced a new variant of EOAC in which edges are identified in HSV color space using a framework based on Full Range Gaussian Markov Random Field (FRGMRF) model that extracts very minute and fine edges from HSV color space and evades loss of edges owing to spectral variations; Gradient field histogram of gradient (GF-HOG) is reported for the retrieval of photo collections [46] and it attains better results than the features like multi-resolution HOG, SIFT, structure tensor, etc. Histograms of triangular regions and relative spatial information for histogrambased representation of the BoVW (Bag of visual words) model are reported in Ali et al [23,24] and Zafar et al [47][48][49] respectively. Feature computation based on spatial information is reported in Zafar et al [47][48][49], Latif et al [50] and Ali et al [51].…”

Estimation of gait normality index based on point clouds through deep auto-encoder

“…Recently, such approaches have demonstrated good performance on common problems such as content-based image retrieval [2,42] and image classification [3,40,41]. Another approach is the use of pretrained deep networks for feature extraction such as [27,31].…”

Mode determination in variational mode decomposition and its application in fault diagnosis of rolling element bearings

“…The extracted features are provided as input to an SVM multi-class classifier utilizing multiple kernel functions to determine their distinguishing capability in classifying faulty and non-faulty bearing signatures. Cross-validation is also performed in SVM to select the best kernel parameters [31][32][33][34].…”