A novel feature based indexing algorithm for brain tumor MR-images
- PDF / 5,544,981 Bytes
- 7 Pages / 595.276 x 790.866 pts Page_size
- 13 Downloads / 176 Views
ORIGINAL RESEARCH
A novel feature based indexing algorithm for brain tumor MR-images Ravindra K. Purwar1 • Varun Srivastava1
Received: 11 September 2019 / Accepted: 6 December 2019 Ó Bharati Vidyapeeth’s Institute of Computer Applications and Management 2019
Abstract A wide use of indexing algorithms for biomedical images is done by the researchers. This paper proposes a novel feature based algorithm for indexing in brain tumor MR images which helps to compare and analyze the extent of disease amongst a database. After the detection of cancer by using various filters and operations, we use features like centroid, length of the parameter, number of connected components and Laplacian of Gaussian for indexing. This enables us to extract similar images in the database as well as to analyze the impact of disease in the unknown sample. Keywords CBIR (content based image retrieval) Image indexing Brain tumor detection
1 Introduction CBIR performs extraction of features from images and compares them with images from a database to procure similarity index. Biomedical image indexing has recently become a field of interest for researchers because of its profound application in medical databases. Polakowski et al. [1] used feature based approach to identify cancer in a mammogram. He used nine features like shape, size etc. to diagnose the cancer affected area. Xu et al. [2] proposed
& Varun Srivastava [email protected] Ravindra K. Purwar [email protected] 1
University School of Information, Communication and Technology, Guru Gobind Singh Indraprastha University, New Delhi, India
the pixel level segmentation to identify the disease and extract the features. The images were then accessed using those features using multiple cluster instance learning. Ojala et al. [3] proposed local binary patterns (LBPs) to extract the texture and thereby retrieve similar images which were extended to local ternary patterns by Tan et al. [4]. Guo et al. [5] also proposed a hybrid approach with a global rotation invariant matching to avoid loss of global spatial information in LBPs. Further Murala et al. [6] proposed local mesh peak valley edge patterns for indexing (LMePVEP) which was found to be much effective for indexing purposes. Vipparthi et al. [7, 8] also enhanced the LBP operators and achieved better results as compared to traditional LBP. Then ternary LMePVEPs were separated into binary ones and the corresponding histograms were concatenated to act as a feature vector. Scott et al., Basin et al. and Akaki et al. [9–11], proposed various image indexing mechanisms based on the knowledge, content and semantics. Unay et al. [12] combined LBP patterns with Kanade–Lucas–Tomasi feature points for identification of brain tumor and compared them with base methods. Murala et al. [13] extracted local mesh patterns and applied Gabor filter on it. It considered the local differences between neighbors and a histogram was thereby computed to form the feature vector. It was based on the technique employed by Vipparthi et al. [14] where the
Data Loading...
