A flexible content‐based image retrieval model and a customizable system for the retrieval of shapes

Abstract

The authors describe a flexible model and a system for content-based image retrieval of objects' shapes. Flexibility is intended as the possibility of customizing the system behavior to the user's needs and perceptions. This is achieved by allowing users to modify the retrieval function. The system implementing this model uses multiple representations to characterize some macroscopic characteristics of the objects shapes. Specifically, the shape indexes describe the global features of the object's contour (represented by the Fourier coefficients), the contour's irregularities (represented by the multifractal spectrum), and the presence of concavities and convexities (represented by the contour scale space distribution). During a query formulation, the user can specify both the preference for the macroscopic shape aspects that he or she considers meaningful for the retrieval, and the desired level of accuracy of the matching, which means that the visual query shape must be considered with a given tolerance in representing the desired shapes. The evaluation experiments showed that this system can be suited to different retrieval behaviors, and that, generally, the combination of the multiple shape representations increases both recall and precision with respect to the application of any single representation.

A new algorithm for product image search based on salient edge characterization

“…Humans can easily distinguish objects of interest by visually examining shape elements present in a digital image. Prior study (Bordogna & Pagani, ) indicated that shape cues can more effectively help the human visual pattern recognition tasks than visual features on the color and texture aspects. Contour‐based shape descriptors and region‐based shape descriptors are two types of shape representations.…”

Questioning the visual: Research methods and the problem of perception

“…These have included studies of the terms applied to images through social tagging (Golbeck, Koepfler & Emmerling, 2011; Sun et al, 2011; Bar‐Ilan, 2010; Rorissa, 2010; Stvilia & Jörgensen, 2010), image description (Zeng, 1999) and image indexing (Ménard, 2013; Ménard & Smithglass, 2012; Wang et al, 2011; Matusiak, 2008; Rorissa & Iyer, 2008; Choi & Rasmussen, 2003; Chen, 2001, Jörgensen, 1998; Fidel, 1997; Enser & McGregor, 1992). Image retrieval studies in the literature within the domain have generally consisted of those that examine automatic retrieval processes based on image content (Bordogna & Pagani 2010; Zachary, Iyengar & Barhen, 2001) and those which rely on text‐based methods (Westman, Laine‐Hernandes & Oittinen, 2011; Greisdorf & O'Connor, 2002; Hastings, 1999). The possible benefit of combining these two methods has also been examined (Apostolova et al, 2013; Névéol et al 2009; Jörgensen, 2003; Enser, 2000).…”