To recognize or not to recognize – A database of encrypted images with subjective recognition ground truth

Low Quality and Recognition of Image Content

“…In addition, we do not know if the VQIs which are available could properly evaluate recognizability, primarily because the relation of quality and recognizability for the HVS in this context is not researched at all. What we do know however, from [18], is that the VQIs most frequently used for this task are not able to perform well in this capacity.…”

“…This is not an optimal situation but at least allows for the development on one database and the testing on the other. For the development of indices which can handle content intelligibility the situation is somewhat more complicated as shown in [18] which also introduces the first database with a recognition score for images encrypted on the border of content recognition. The database presented in [17] also contains a recognizability score (denoted content leak information), which is directly rated by observers when comparing the original to the encrypted image, in addition to the visual quality data.…”

“…The presentation of the original coupled with the tendency of the human visual system to find patterns strongly influences the content leak information. The translation of this content leak information [17] to the recognition index [18] is difficult at best. Specifically, both are susceptible to pareidolia, i.e., the tendency for incorrect perception of a stimulus as an object, but the method in [18] is designed to catch that (multiple originals) while the setup in [17] is not (single original).…”

“…The translation of this content leak information [17] to the recognition index [18] is difficult at best. Specifically, both are susceptible to pareidolia, i.e., the tendency for incorrect perception of a stimulus as an object, but the method in [18] is designed to catch that (multiple originals) while the setup in [17] is not (single original). The data in [17] is still useful for the security evaluation of encrypted images, but not the evaluation of the recognition threshold.…”

“…In this paper we present a subset of the database from [18] to a panel of human observers to get an evaluation of perceived quality for encrypted images at the recognition threshold. This is not only a further database for the development of quality based indices for selective encryption but also allows us to study the differences and commonalities between quality and recognizability, a hitherto unstudied subject.…”