If you mean the receptive field characteristics or representations of visually activated neurons, there is a lot of studies showing that top down attention, enhances this activity and makes it more persistent (Desimone, Reynolds, Treue, Maunsell, Shadlen, Roelfsema, etc). ergo enhances and extends its storage. Short term memory processes are thought to be mediated by persistent neural activity. Information is stored where it is represented by activated neurons.
While my relevant EEG data are under review, I would say such information is stored for at least 1 second. The "filtered" information in question concerns features of distractors that we are not aware of, whilst we are aware of the presence of those distractors that we ignore. This view concerns EEG data measured from occipital sites as could be consistent with van der Togt's response above, with the caveat that localisation of EEG source generation is plagued by the electromagnetic inverse problem.
To Chris: let me explain, by saying "filtered information", I mean those data that are excluded by attention, specially at the level of V1. From our entire visual perception, attention allows some data to pass through v1 to higher level like MT, IT ,on the basis of visual deployment. it means some information will be saved, so what happens to not entered data ?
To Tom. I suppose you talk about iconic memory. But I am looking for something beyond the iconic memory, something between iconic memory and long term one.
Partial report demonstrates that items held in what has been termed iconic memory can be reported, a subset at a time. Report was not possible for multiple surprise tests of memory of features for the ignored distractor, either on the first or subsequent surprise tests. The brain's memory for features of distractors lasts at least a second, not less, maybe more.
Sure, if we get to that stage, I'll send you the proofs. The partial report work was done by George Sperling in the 1960s. http://aris.ss.uci.edu/HIPLab/staff/sperling/PDFs/Sperling_PsychMonogr_1960.pdf
To come back to your question of not entered data. I would like to mention that many researchers now believe that attentional processes are mediated by feedback between lower and higher visual areas. In other words features and representations come into the attentional domain when recurrent activity develops between higher and lower areas for particular features. It's possible that this recurrent activity is the mediator of conscious working memory (see model of Dehaene). So visual input that leads to activation of neurons up to the prefrontal cortex, e.g Frontral eye fields, probably involves phasic responses within 100ms and die out after 200ms, if they do not attract or engage our attentional systems. Attentional modulation even down to V1 operates mostly after 200ms.
Interestingly these unconsciously activated features can later be brought in to attention (iconic memory) if there has not been new visual input causing backward masking. The reason is probably that neural activity lingers on and becomes suppressed only after the activation of other neurons.
As a follow on from Chis' point. The key seems to be reverberation (see the details to the enclosed papers), which is faciliatated/mediated by the attention process. Mojtaba, you might find these little papers at the following web location of use in understanding the processes:
(1) The first one looks at some of the key mechanics, and is at: http://www.seahorses-consulting.com/DownloadableFiles/Appendix1.pdf.
(2) The second one looks at key visual attention processes, and puts these into context. Here is the link: http://www.seahorses-consulting.com/DownloadableFiles/ShapingAttentionHandout.pdf