Biomimicry of Human Pattern Recognition by Puzzle Solving Simulation
No Thumbnail Available
Date
2023
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
none
In this work, our purpose is to imitate human behavior in pattern recognition by puzzle solving simulation with an automatic algorithm based on statistic database of human solver. Based on the empirical database of puzzle solving of 972 human solvers, it has been observed that human solvers tend to pick a piece as the nucleation site and then enlarge the site by finding out corresponding piece of its edges with similar color pattern. In this study, an automated algorithm has been developed based on the empirical data from the previous research. The algorithm incorporates specific parameters that are crucial for puzzle solving, including the number of sections for each puzzle piece, the resemblance threshold, alpha, the percentage of ABC, and q values. The objective of the study is to evaluate the simulation performance by comparing it with the empirical data for different parameter settings. Our simulation shows that by setting the Number of sections into 6×6, Resemblance threshold 0.65, Alpha 0.55, q values 5, and Percentage of ABC {90,8,2}, our simulation that working based on color does mimics human solvers with strong effect size r^2 0.72 for 6 Pictures that dominates by colors. At the second measurement, we found that the simulation with number of sections 6×6, Resemblance threshold 0.65, Alpha 0.55, q values 1, and Percentage of ABC {94,4,2} showcased the best performance, with R-squared value of 0.82 and a Spearman's correlation coefficient of 0.85 for the set of 8 pictures. Similarly, for the set of 6 pictures, it exhibited an R-squared value of 0.87 and a Spearman's correlation coefficient of 0.94.
In this work, our purpose is to imitate human behavior in pattern recognition by puzzle solving simulation with an automatic algorithm based on statistic database of human solver. Based on the empirical database of puzzle solving of 972 human solvers, it has been observed that human solvers tend to pick a piece as the nucleation site and then enlarge the site by finding out corresponding piece of its edges with similar color pattern. In this study, an automated algorithm has been developed based on the empirical data from the previous research. The algorithm incorporates specific parameters that are crucial for puzzle solving, including the number of sections for each puzzle piece, the resemblance threshold, alpha, the percentage of ABC, and q values. The objective of the study is to evaluate the simulation performance by comparing it with the empirical data for different parameter settings. Our simulation shows that by setting the Number of sections into 6×6, Resemblance threshold 0.65, Alpha 0.55, q values 5, and Percentage of ABC {90,8,2}, our simulation that working based on color does mimics human solvers with strong effect size r^2 0.72 for 6 Pictures that dominates by colors. At the second measurement, we found that the simulation with number of sections 6×6, Resemblance threshold 0.65, Alpha 0.55, q values 1, and Percentage of ABC {94,4,2} showcased the best performance, with R-squared value of 0.82 and a Spearman's correlation coefficient of 0.85 for the set of 8 pictures. Similarly, for the set of 6 pictures, it exhibited an R-squared value of 0.87 and a Spearman's correlation coefficient of 0.94.
Description
Keywords
none, Puzzle solving, Algorithm, Pattern recognition, R-squared, Spearman's correlation