4
Abstract
With the advent of solutions provided to science by computers, automation solidified
its place as the basis of performance. Through automated solvers, dynamic problems of
increased complexity and changing nature are solved. Although solvers provide solutions in
such problems, studying and optimizing them brings new questions to the surface. One of the
challenges is solving games as their extensive studies leads to complex computing problems.
Utilizing a medium such as games facilitates the observation of solver results. As a result,
there have been many milestones in the history of game solving.
In the present diploma thesis, the game of Minesweeper is examined and a unique
approach is implemented with the aim of extending pre-existing methods of visual parsing
and solving. The Automated Solver combines machine vision techniques to find the board and
identify the state of individual blocks. At the same time, it utilized deterministic algorithms
for simple board states while modeling the more complex ones as Constraint Satisfaction
Problems (CSP). This model is input in a CP-SAT solver based on the hybrid LCG technique,
to be encoded and solved by the internal SAT solver with the corresponding computational
advantages. Based on the solutions provided by CP-SAT for possible mine assignments, a
possibility analysis is performed on the board giving precision to the next moves.
The implementation of the diploma thesis presents both the advantages of visual
parsing and solution as a CSP, as well as the limitations and challenges involved. At the same
time, through an experimental process, the correlations between metrics and the influence of
mine density are analyzed.
Keywords:
Machine vision, digital image processing, image parsing, constraint satisfaction
problems, C#