Abstract:

Previous studies have found that perceptual learning can improve the performance on visual search tasks. However, many cognitive processes are involved in visual search, and it is unclear at which visual processing stage perceptual learning facilitates search performance. The current study explored the mechanism of perceptual learning by dividing the eye movement metrics into three visual processing stages: search initiation time (the early visual processing stage), which represents the cognitive process of the time of processing the current location and selecting the next search location; scanning time (the middle visual processing stage), which includes the number and processing time of fixation positions; verification time (the late visual processing stage), which represents determining whether the current stimulus is the target and making a verification. A 2 (target type: trained vs. untrained triangle) × 2 (test stage: pretest vs. posttest) within-subjects design was used to address the above issue. 24 healthy young adults (5 males; mean age: 21.23 ± 2.02 years) participated as paid volunteers in this study. We randomly selected one of the four orientations of the triangle (Up, Down, Left, or Right) as the trained triangle, which would receive three days of training. Moreover, to ensure that the visual search training was specific to the trained triangle, the trained and untrained triangles were tested by recording the behavior results and eye movement before and after training (untrained triangle was randomly selected from the distractors). Each trial started with a fixation cross (When eye movement was recorded, the search display would not appear until the participants fixated on the center cross for more than 500 ms; when eye movement was not recorded, the central fixation cross was presented for 500 ms and then the search screen was presented). Then a search display was presented until the key response or the elapse reached 2000 ms since its onset. The response was recorded only before the fixation cross disappeared. The task of participants was to determine whether or not the target was presented as quickly as possible. Participants pressed the left arrow key to report the presence of a target or the right arrow key to report its absence. A two-way repeated-measures ANOVA was conducted with the factors of target type (trained vs. untrained triangle) and test stage (pretest vs. posttest). The behavior results (Figure 1) found the reduced response time (target present trial: Δ 0.50 ± 0.10 s, t(23) = 24.26, p < 0.001, Cohen’s d =5.06, BF₁₀ = 7.29, 95% CI = [0.45, 0.54]; target absent trial: Δ 0.45 ± 0.20 s, t(23) = 11.06, p < 0.001, Cohen’s d = 2.28, BF₁₀ = 8.74, 95% CI = [0.36, 0.53]) and increased accuracy (Δ -0.37 ± 0.14, t(23) = -13.31, p < 0.001, Cohen’s d = 2.77, BF₁₀ = 2.99, 95% CI = [-0.43, -0.31]) when searching for trained stimuli after training. However, there was no significant difference in response time (target present trial: Δ -0.01 ± 0.18 s, t(23) = -0.40, p = 0.696; target absent trial: Δ 0.04 ± 0.17 s, t(23) = 1.13, p = 0.270) or accuracy between pretest and posttest for untrained stimuli(Δ 0.00 ± 0.18, t(23) = 0.07, p = 0.942). The results of eye movement tracking are as follows: (1) in the early visual processing stage (Figure 2), the search initiation time of the trained stimuli increased significantly after training(target present trial: Δ -32.43 ± 63.95 ms, t(23) = -2.48, p = 0.021, Cohen’s d = 0.52, BF₁₀ = 2.65, 95% CI = [-59.43, -5.42]; target absent trial: Δ -45.16 ± 75.56 ms, t(23) = -2.93, p = 0.008, Cohen’s d = 0.61, BF₁₀ = 6.12, 95% CI = [-77.06, -13.25]), and there was no significant difference in the search initiation time between pretest and posttest for untrained stimuli(target present trial: Δ 0.31 ± 83.42 ms, t(23) = 0.02, p = 0.986; target absent trial: Δ 13.51 ± 101.67 ms, t(23) = 0.65, p = 0.52). (2) In the middle visual processing stage, the number of fixations of trained stimuli (Figure 3) were significantly reduced after training(target present trial: Δ 2.04 ± 0.50, t(23) = 19.89, p < 0.001, Cohen’s d = 4.15, BF₁₀ = 9.37, 95% CI = [1.83, 2.26] ; target absent trial: Δ 2.23 ± 0.85, t(23) = 12.84, p < 0.001, Cohen’s d = 2.68, BF₁₀ = 1.46, 95% CI = [1.87, 2.59]) and there was no significant difference for untrained stimuli before and after training(target present trial: Δ -0.02 ± 1.16, t(23) = -0.10, p = 0.919 ; target absent trial: Δ 0.18 ± 1.14, t(23) = 0.79, p = 0.437). The average fixation time of trained stimuli (Figure 4) was significantly reduced after training when target present (target present trial: Δ 63.40 ± 42.04 ms, t(23) = 7.39, p < 0.001, Cohen’s d = 1.54, log (BF₁₀ ) = 11.47, 95% CI = [45.65, 81.16]), but there was no significant difference for untrained stimuli before and after training(target present trial: Δ -4.82 ± 23.23 ms, t(23) = -1.02, p = 0.321). There was only a significant main effect of test stage in average fixation time when target absent (target absent trial: F (1, 23) = 10.06, p < 0.01, η2 p = 0.30). (3) In the late visual processing stage, there was no significant difference in verification time (Figure 5) between the pretest and posttest for both trained and untrained stimuli (target present trial: F(1, 23) = 1.25, p = 0.274; target absent trial: F(1, 23) = 0.37, p = 0.552). In conclusion, the accuracy and search initiation time of searching for trained stimuli was increased, while the number of fixations and the fixation time decreased. Moreover, the changes in behavior and eye movement indexes did not transfer to untrained stimuli. It is suggested that perceptual learning can affect the early and middle visual processing stages, and search performance may be improved by increasing the search latency, reducing the number of saccades, and reducing the fixation time.

Key words: perceptual learning, nonsalient stimuli, learning mechanism, visual processing, eye movement