Efficient descriptors selection in automatic image retrieval using DENOL

Issue title: Soft Computing Applications

Guest editors: Valentina Emilia Balas

Article type: Research Article

Authors: Ignat, Anca^a | Luca, Mihaela^{b; *} | Păvăloi, Ioan^b | Lazăr, Camelia^b

Affiliations: [a] Faculty of Computer Science, University “Alexandru Ioan Cuza” of Iaşi, Iaşi, Romania | [b] Institute of Computer Science, Romanian Academy Iaşi Branch, Iaşi, Romania

Correspondence: [*] Corresponding author. Mihaela Luca and Ioan Păvăloi, Institute of Computer Science, Romanian Academy Iaşi Branch, Iaşi, Romania. E-mail: mihaela.luca@iit.academiaromana-is.ro (Mihaela Luca); E-mail: ioan.pavaloi@iit.academiaromana-is.ro (Ioan Păvăloi).

Abstract: A well-structured and indexed database alleviates the computing burden on large data. This paper describes groundwork for presenting the data in a compact, distinctive form, improving the procedures of applying keypoint detection algorithms to preprocess and reduce the relevant features of the images. Our method computes for an image a number of SURF keypoints in a given interval, by adapting the threshold related to the Hessian matrix blob detector. This type of approach allows selecting the level of detail to use in image description and gives us control over the computing time. We named this method DENOL (Descriptor Number On Limits) and tested it on images from two datasets, UCID and an original image database which we propose, IIT_DB. Very good retrieval results and a significantly reduced computing time are achieved.

Keywords: CBIR, keypoint detector, keypoint matching, image database, labeling, UCID, IIT_DB

DOI: 10.3233/JIFS-219275

Journal: Journal of Intelligent & Fuzzy Systems, vol. 43, no. 2, pp. 1739-1749, 2022