Study And Implementation Of Determinding The Outcome Based On The Final Go Image

Over the past two years,the emergence of the artificial intelligence AlphaGo has raised a new round of fever in Go.The offline competition of Go has become more and more popular,which produces great request to the referees.During the current offline Go,the competition results are always determined by the conventional method where it is done by manual.With complicated steps,the manual way costs too much time and it doesn't guarantee to achieve a satisfactory result.To address the above problems,this thesis proposes a counting algorithm for offline Go,which is based on digital image processing.The algorithm handles the final image of the competition and gives the result automatically.The proposed algorithm is capable of dealing with different sizes of checkerboard,different rules of games and some complicated cases,such as captured stones and impasse.Compared with the way by manual,the proposed method takes less time on counting,reduces more burden of the referees,and achieves a more accurate result.The aim of the proposed method is that it provides a convenient and prompt way to determine the Go competition result.For go game final image recognition and analysis,in the beginning we fetch the final images of three different board sizes including 19*19,13*13 and 9*9 boards,operate these images with perspective transformation and bilinear interpolation processing to get information only within the board outline and perspective images directly above the board.Then the gray and red components of the preprocessed image are extracted,split the black pieces.The blue component of the preprocessed image is extracted and the white pieces are segmented.Comparing the three different threshold segmentation method including 1)histogram threshold segmentation method,2)iterative threshold segmentation method,and 3)the between-cluster variance threshold segmentation method,get the final segmentation result through morphological processing during the partition process.The results show that the red component of preprocessing the image is more suitable for the segmentation of black stones,and the histogram threshold segmentation method can more accurately identify and split black stone and white stone.The location of all the intersections in the board can be found byprojection.According to the location of all these intersections we can detect the position of all black and white stones,and fill in the two-dimensional arrays.In the final judgement,considering three different rules including China,Japan and South Korea rules,I design three corresponding method of processing the captured stones,complex double living situation of judgement and the rest of the intersection of filling.In the final stage,the total number of pieces under the corresponding rules is calculated,and the result is determined according to the rules which is preset.Among them,the Chinese rule is similar to the Ying's rules,so after finishing filling the rest of the intersection,we only need to count the totals statistics.The Japanese and Korean rules need to count the number of remaining intersections,as well as the number of dead and takes.The results show that all three methods are accurate and feasible.