Solving pictorial jigsaw puzzles via Internet-based collective in-telligence

Abstract

Pictorial jigsaw (PJ) puzzles are a well-known leisure game for humans. Usually, a PJ puzzle game is played by one or several players face-to-face in physical space. In this paper, we focus on how to solve PJ puzzles in cyberspace by a group of physically-distributed players. We propose an approach to solve PJ puzzles using stigmergy-inspired Internet-based human collective intelligence. The core of the approach is a continuously- executing loop, namely, the EIF loop, which consists of three activities: exploration, integration, and feedback. During the exploration activity, each player tries to solve the PJ puzzle alone, without direct interaction with the other players. At any time, the result of a player's exploration represents a partial solution to the PJ puzzle and a set of rejected neighboring relationships between the puzzle pieces. The results of all players' explorations are integrated in real time through the integration activity, forming a continuously-updated collective opinion graph (COG). Through the feedback activity, each player is provided with personalized feedback information based on the current COG and the player's current exploration results, in order to accelerate his/her puzzle-solving process. Ex- ploratory experiments show that: (1) Supported by this approach, the time to solve PJ puzzles is nearly linear with respect to the reciprocal of the number of players. Furthermore, compared with the best single players in the ex- periments, the puzzle-solving time decreases by 31.36%{64.57% on average for groups composed of 2 to 10 players. (2) Supported by this approach, the feedback information received by the best player in a group has an average precision of 86.34%, and, as the group size increases, the feedback information in the puzzle-solving result of the best player increases gradually from 20% to 45%. (3) This approach exhibits a better scalability with puzzle size than that of a face-to-face collaboration among ten players. Additionally, it always leads to 100%-accurate puzzle- solving results, whereas the results of the automated PJ-puzzle solver have an average accuracy of only 52%. It is envisaged that these results will provide useful information or opinions to facilitate a broader application of human collective intelligence in the Internet environment.