Transfer Generalization Of Contour Integration Learning In Path And Orientation Dimension

Contour integration,the way by which the visual system groups discrete elements in cluttered background into a whole continuous contour,is a critical early step that bridges primary sensory processing and higher-level object-based perception under the natural environment.Although evolution and development have made the visual system be expert in contour integration processing,its cortical plasticity is still largely unknown.Perceptual learning refers to a long-term and stable enhancement of perceptual task performance as a result of perceptual experience.An important tool used as a behavioral probe to assess the cortical plasticity is characterization of learning specificity and transfer.Specificity refers to the phenomenon that the improvement will lost when the stimulus or location is changed,whereas transfer means other untrained stimuli or locations can be improved after training.The investigation of specificity and transfer of learning effect can provide behavioral evidence for the cortical plasticity which induced by perceptual learning.Previous studies have demonstrated that contour integration can be improved by short-term perceptual training.However,it still remains unclear whether the learning effect can transfer across different contour dimensions.To clarify this question,we examined the specificity and transfer in contour path and contour type in Experiment 1.Contour type was defined by the grouping regularity of contour elements.To be specific,the type that the contour elements aligned along the invisible tested contour path was collinear contour,perpendicular to the path was orthogonal contour,and other cases were acute contour.Contour path referred to the shape of the invisible line(e.g.straight,curvature).Each experiment had four phases: pre-test,training,post-test I,and post-test II.In Experiments 1A & 1B,participants were trained to detect orthogonal-straight contours and orthogonal-curvature contours embedded in Gabor field during the training phase,respectively.In test phases,four contour conditions(2(contour types: collinear,orthogonal)× 2(contour paths: straight,curvature))were tested in each experiment.We observed significant improvement for trained conditions and the untrained conditions which shared the same contour type but different contour path with trained condition after training in both Experiment 1A and 1B.However,untrained conditions that had different trained type with trained conditions were failed to produce significant improvement after training.In addition,there were no significant difference between post-test I and post-test II in both two experiments,which indicated the enhancement was stable.To exclude the impact of task difficulty between different contour types,a new contour type(acute contour)which matched difficulty with orthogonal contour before training by pilot,was added in Experiment 2.Experiment 2also included two training groups: orthogonal-straight contour training group(Experiment 2A)and orthogonal-curvature contour training group(Experiment 2B).Moreover,in order to ensure that the thresholds of trained contour in two groups could reach to an asymptotic level,we prolonged training sessions to 7 days.Six contour conditions(3(contour types: collinear,orthogonal,acute)× 2(contour paths: straight,curvature))were tested in test sessions of each experiment.Consistent with Experiment1,we found that the learning effect can be transferred to the untrained contour path of trained type,but not to the untrained contour type of trained or untrained contour path.Combined the results of Experiment 1 and Experiment 2,our research showed the type specificity of contour integration learning.More importantly,we found that learning effect could be transferred to the untrained condition which had the same contour type but different path with trained condition,which indicated that contour integration learning was transferable between different contour paths.In order to examine the generalization of the results of Experiment 1 and Experiment2,we added a new dimension named global orientation in Experiment 3.To be specific,one-half of participants in Experiment 3A(orthogonal-straight contour training group)and Experiment 3B(orthogonal-curvature contour training group)were trained with contours whose global orientation were near 45°(±15°),whereas the rest were trained with contour orientation near 135°(±15°).Four contour conditions(2(contour paths:straight,curvature)× 2(contour orientation: trained,untrained))were tested in test sessions.The results showed significant transfer of learning effect between contour paths,which were consistent with Experiment 1 & 2.Furthermore,we also found significant improvement of untrained contour orientations,which indicated contour integration learning was also transferable between different contour orientations.In conclusion,contour integration learning showed two important characteristics of perceptual learning:specificity and persistence.What’s more,we found that contour integration learning was transferrable between different contour paths and global contour orientations.These results suggested that short-term experience could improve the individuals’ ability to detect all contour paths with trained grouping regularity in cluttered background,which reflected the flexibility of visual system to exploit grouping regularity in natural scenes and the key role of experience.Our findings provided new behavioral evidence for understanding the mechanism of cortical plasticity in adult.