In this work, we address the jigsaw puzzle solving task, proposing an automated pipeline to assess the adjacency relationship among tiles and order them. In particular, we compare two approaches Relaxation Labeling (ReLab) and Puzzle Solving by Quadratic Programming (PSQP). We train convolutional neural networks (CNNs), trying different methods to extract compatibility between tiles of images, first by approaching the task as a supervised learning problem and then by using self-supervised learning, a variation of the unsupervised learning theme. We build a CNN trained for a pretext task, which can later be repurposed to extract tiles compatibility. Finally, we test different combinations of CNNs -- as automatic feature extractors -- and puzzle solving methods on publicly available datasets, providing the feasibility of our proposed solution.
Using residual neural networks for jigsaw puzzle solving.
Zonelli, Mattia
2022/2023
Abstract
In this work, we address the jigsaw puzzle solving task, proposing an automated pipeline to assess the adjacency relationship among tiles and order them. In particular, we compare two approaches Relaxation Labeling (ReLab) and Puzzle Solving by Quadratic Programming (PSQP). We train convolutional neural networks (CNNs), trying different methods to extract compatibility between tiles of images, first by approaching the task as a supervised learning problem and then by using self-supervised learning, a variation of the unsupervised learning theme. We build a CNN trained for a pretext task, which can later be repurposed to extract tiles compatibility. Finally, we test different combinations of CNNs -- as automatic feature extractors -- and puzzle solving methods on publicly available datasets, providing the feasibility of our proposed solution.File | Dimensione | Formato | |
---|---|---|---|
870038-1270403.pdf
accesso aperto
Tipologia:
Altro materiale allegato
Dimensione
4.88 MB
Formato
Adobe PDF
|
4.88 MB | Adobe PDF | Visualizza/Apri |
I documenti in UNITESI sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/20.500.14247/15736