In this study, a system is designed to reconstruct multi-fragmented 2D objects with faster and less errors with the computer aided. In the proposed system, the reference images related to fragmented objects are used. Since the reconstruction process of fragmented objects contains quite challenging sub-problems, it is divided into sub-stages and various contributions are made to the literature. In the first stage, since there are no datasets on the internet, a comprehensive dataset is created. At the stage of matching the pieces with reference images, the keypoint based SIFT, SURF, BRISK and AKAZE methods are compared. Owing to the proposed Borda count based approach, AKAZE method is decided to use at the matching stage. Also, a block based method that used Zernike moments is developed for sub-problems where keypoint based methods are insufficient. An alternative method based on minimum convex hull is developed for the problem of alignment of pieces. At the determination of the pixels of the pieces, the CIELAB color space based approach is proposed to overcome the color problems arising from the scanners. Global consistency are achieved by approach that takes into account the edges and color information similarities of the pieces at the multiple reconstruction process. In addition, a method that grouped the pieces of the 2D objects within themselves is proposed for the first time and it is shown that the problem solving can be accelerated by grouping process. |