Table 1.

Methods implementing image segmentation in IR

Method	Description	Limitations
Thresholding or Binarization (75,76)	This is a method based on the image segmentation, which create binary of gray scale images to perform image analysis. Various methods (e.g. point dependent techniques, region dependent techniques, local thresholding, multithresholding, Histogram Thresholding (77), Picture Thresholding (78), minimum spatial entropy threshold (79), Fuzzy entropy thresholding (80) etc.) have been proposed for thresholding.	Incorrectly set threshold can lead to under or over segmentation of objects (75).
Clustering	To understand large-scale complex data (text and images etc.), this method is widely applied in different fields (e.g. information retrieval, bioimaging, medicine etc.) for pattern recognition, speech analysis and information retrieval (81). To perform image features and text analysis, clustering divides content in possible numbers of meaningful group objects by breaking it into subcategories and draw relationships between them (82). There have been many different methods [e.g. Image segmentation by Clustering (83), Dual clustering (84) etc.], techniques (K-means clustering, Hierarchical clustering, Partitional clustering, Exclusive Overlapping clustering, Fuzzy clustering, Fuzzy C-means (FCM) clustering, Complete clustering, Partial clustering, Agglomerative Hierarchical Clustering, etc.) and types (Well-Separated, Prototype-Based, Graph-Based, Density-Based, Shared-Property etc.) for clustering.	It is difficult to predict fixed number of clusters while grouping objects and it consumes extensive computational time.
High Dimensional Indexing (HDI) (85)	There have been many HDI techniques proposed for large scaled content-based image retrieval, which have been categorized in Dimension Reduction [embedded dimension, Karhunen–Loeve transform (KLT), low-rank singular value decomposition (SVD) etc.], and Multi-dimensional indexing (Bucketing algorithm, priority k-d tree, quad-tree, K-D-B tree, hB-tree, R tree etc.) techniques (86).	Blind dimension reduction might not bring optimistic results during embedded dimension reduction.