This document describes the Compact Descriptors for Visual Search (CDVS) Test Model (TM) 13.0 (CDVS1) (CDVS2). In particular, this document looks at the TM architecture, the operation of the informative component technologies (descriptor matching and retrieval) of the TM, and the offline or training procedures for both the normative (descriptor extraction) and the informative (descriptor matching and retrieval) component technologies of the TM. This document should be read in conjunction with the Text of ISO/IEC FDIS 15938-13 Compact Descriptors for Visual Search (CDVS3), which examines the normative (descriptor extraction) pipeline.
The TM implements the required functionality for the extraction and comparison of compact descriptors constrained to a set of predetermined descriptor lengths.
In particular, two procedures for descriptor comparison are implemented in TM, aiming at reproducing two fundamental tasks for real visual search systems: pairwise matching and retrieval. The former regards automated verification of whether two images depict the same objects or scene; in this case, descriptors extracted from a query image are matched against the descriptors of a reference image, in order to determine whether they match or not. The latter regards the search and discovery of images contained within a large collection that depict the same objects or scenes as those depicted by a query image; this requires the database images to be processed for the creation of a database which may be searched using the descriptors extracted from the query.
The current document is structured as follows: Section 2 briefly looks at the extraction of compact descriptors for visual search. Sections 3 and 4 describe the operations performed by the TM for the pairwise matching and retrieval procedures, respectively. Section 5 provides some insight about the reference software structure and architecture. Some of the operations supported by TM require training: the procedures enabling such a training phase are described in the Appendices.