Integration of bottom-up and top-down processes in visual object recognition Annette M. Schmid,...
-
Upload
milton-roberts -
Category
Documents
-
view
224 -
download
0
Transcript of Integration of bottom-up and top-down processes in visual object recognition Annette M. Schmid,...
Integration of bottom-up and top-down processes in visual object Integration of bottom-up and top-down processes in visual object recognitionrecognition
Annette M. Schmid, Marianna Eddy, Phillip J. HolcombDepartment of Psychology, Tufts University, Medford, MA
IntroductionNumerous anatomical studies demonstrate strongly bidirectional connections among cortical (and subcortical) areas (e.g. Barbas 2000; Onguer & Price 2000). The importance of these anatomical findings has been highlighted by the theoretically posed need of integrated top-down and bottom-up processes in efficient (e.g. Bullier 2001; Mumford 1993) visual processing of complex stimuli.In addition functional experiments reveal evidence of modulations of early processing areas that can best be explained by top-down influences. (e.g. Rees et al. 2001). One way of looking at integrated models is in the framework of contingency models, where early information, such as global aspects of an object, constrain the evaluation of possible interpretations of an input (Sanocki, 1993).
Hypothesis and Predictions
Strong priming effect in repetition condition (irrespective of behavioral task)A strong priming effect following full bandwidth (intact) primesEarly frontal ERPs that are congruent with a prefrontal source of top-down facilitationDifferential repetition priming effects with different frequency compositions
• Possible prefrontal with intact and low-spatial frequency primes• Possible posterior temporal with intact and high-spatial frequency primes
Reaction time advantage for intact mechanism
Electrophysiological results:
We are particularly interested here in how the bottom-up and top-down processes might be integrated. Bar proposed in 2003 (see diagram 1) a model that emphasizes bidirectional information flow of the cerebral cortex (Ullman, 1995; Sanocki, 1993) and aims to explain how a top-down facilitation might be triggered in object recognition (Bar, 2003). Specifically, he suggested that rapid projection of a crude (low-spatial frequency) image to the prefrontal cortex, causes early sensitization of the most likely interpretations of the input image, which are then integrated with the bottom-up processes. The subsequent integration of these fewer candidate objects with the detailed (and slower) bottom-up analysis would then facilitate recognition by substantially limiting the amount of stored object representations that need to be considered (Bar, 2003; Schmid & Bar, 2002; Sanocki, 1993). If this processing represents the general mode of object recognition in everyday life, one might expect it to be automatic. The following experiments therefore address specifically automatic perceptual processes.
Diagram 1
Priming effects expressed by difference waves:
Repetition priming effects at FP1:(Diagram 7)
priming w. intact priming w. low-pass priming w. high-pass
Localization of sources of difference waves:
90 ms
90 ms
160ms 328ms
143 ms 200 ms 328 ms
time
intact prime
low-spatialfrequency prime
Target following
Diagram 11
Diagram 12
Experiment 1 (Diagram 6)
Experiment 2
Repetition priming effects
prime targetprime = target
p<0.001
* *
D 265
References:Bar M. (2003) A cortical mechanism for triggering top-down facilitation in visual object recognition. J. Cogn. Neuroscience 15: 600-609.Barbas H. (2000) Connections underlying the synthesis of cognition, memory, and emotion in primate prefrontal cortices. Brain Res. Bull. 52: 319-330.Bullier J. (2001) Integrated model of visual processing. Brain Res. Reviews 36: 96-107.Ullman S. (1995) Sequence seeking and counter streams: a computational model for bidirectional information flow in the visual cortex. Cer. Cortex 1:1-11.Onguer D., Price J.L. (2000) The organization of networks within the orbital and medial prefrontal cortex of rats, monkeys and humans. Cer. Cortex 10: 206-219.Pascual-Marqui R.D., Esselen M., Koch K., Lehmann D. (2002) Functional imaging with low-resolution brain electromagnetic tomography (LORETA): a review. Meth. Find. Exp. Clin. Pharmacol. 24 Suppl C: 91-95.Rees G., Frith C. Lavie N. (2001) Processing of irrelevant visual motion during performance of an auditory attention task. Neuropsychologia 39:937-947.Schmid A., Bar M. (2002) Selective involvement of the prefrontal cortex in visual object recognition. Soc. Neurosci. M., Orlando, Fl.Schmid A., Bar M. (2003) Activation of multiple candidate object representation during top-down facilitation of object recognition. Soc. Neurosci. M., New Orleans, Li.Sanocki T. (1993) Time course of object identification:evidence for global-to-local contingency. J. Exp. Psychol. Hum. Perc. Perf. 19: 878-898.
Methods
•27 subjects (20 +/- 1.7 years) 9 males,•Fully balanced, 3 x 2 factorial design (see diagram 3), pseudo-randomized design•53 items per conditions
Standard 32 electrode montage (Electrocap International, see diagram 4)
Signal was strengthened by bio-amplifier and bandpassed at 0.01- 40 Hz
Sampling rate 200 Hz
In-house tools were used for stimulus presentation and main analysis (grand average, mean amplitude, peak latency and difference waves)
Localization analysis with LORETA (Low Resolution Brain Electromagnetic Tomography; Pasqual-Marqui et al. 2003)
Analysis
Diagram 4
Discussion and Conclusions
prime targetprime= target
p<0.001
* *
Diagram 10
Repetition priming effects at FP2:(Diagram 8)
Repetition priming effects at O2:(Diagram 9)
Spatial frequency effects:
p<0.001 for mean amplitude and peak latency, significant for all 3 conditions
intact prime low-pass prime high-pass prime
The behavioral data suggest that for a behavioral advantage to be observed the full integration of bottom-up and top-down is necessary.
The time frame of data presented here highlights perceptual rather than semantic effects.
There are significant differences in the early time window following primes of different spatial frequency compositions.
Anterior differences suggested specific effects following primes containing low-spatial frequencies.
Posterior waveforms indicate additional processes following priming with stimuli containing low-spatial frequencies.
While stimuli are matched and balanced across conditions we found strong priming effects are acting on the early visual areas, suggesting a strong top-down modulation.
Although resolution of the LORETA localization is limited, timeframe, characteristics and localization of the early ERPs observed, and the anterior activity in particular, are consistent with the proposed model of top-down facilitation.
Experiment 1: Is there a food item?
Experiment 2: Is this a “real” object or an abstract sculpture?
Repetition Priming No Priming
“intact” prime
“low-spatial frequency” prime
“high-spatial frequency” prime
primes primestargets targets
•16 subjects (20 years) 7 males, •Fully balanced 3 x 2 factorial design (only two conditions shown here, Diagram 2), pseudo-randomized design•40 items per condition
Debriefing suggested that subjects were unable to recognize primes
10 % of trials rejectedMean accuracy 93 %0.9 % real object not recognized6.1% abstract sculpture taken for real objectRepetition priming with intact images is
significantly different to all other conditions (p<0.001, see diagram 5)
Behavioral results
*
*
prime targetprime= target
p<0.001
**
Diagram 2
Diagram 3
500
520
540
560
580
600
620
sameintact
same lp same hp differentintact
differentlp
differenthp
reac
tio
n t
ime
(ms)
Diagram 5
*