Skip to Main content Skip to Navigation

Graph representation and mining applied in comic images retrieval

Abstract : In information retrieval tasks from image databases where content representation is based on graphs, the evaluation of similarity is based both on the appearance of spatial entities and on their mutual relationships. In this thesis we present a novel scheme of Attributed Relational Adjacency Graphs representation and mining, which has been applied in content-based retrieval of comic images. The images used in this thesis are comics images, which have their inherent difficulties in applying content-based retrieval, such as their abstractness, partial occlusion, scale change and shape deformation due to viewpoint changes. We propose a graph representation that yields stable graphs and allow to retain high-level and structural information of objects of interest in comic images. Next, we extend the indexing and matching problem to graph structures representing the comic image, and apply it to the problem of retrieval. The graphs in the graph database representing the whole comic volume are then mined for frequent patterns (or frequent substructures). This step is to overcome the non-repeatability problem caused by the unavoidable errors introduced into the graph structure during the graph construction stage, which ultimately create a semantic gap between the graph and the content of the comic image. Finally, we demonstrate the effectiveness of the system with a database of annotated comic images. Experiments of performance measures is addressed to evaluate the performance of this CBIR system.
Complete list of metadata

Cited literature [287 references]  Display  Hide  Download
Contributor : ABES STAR :  Contact
Submitted on : Wednesday, February 12, 2020 - 10:46:07 AM
Last modification on : Friday, June 3, 2022 - 10:24:28 AM
Long-term archiving on: : Wednesday, May 13, 2020 - 2:25:28 PM


Version validated by the jury (STAR)


  • HAL Id : tel-02475609, version 1



Thanh Nam Le. Graph representation and mining applied in comic images retrieval. Computer Vision and Pattern Recognition [cs.CV]. Université de La Rochelle, 2019. English. ⟨NNT : 2019LAROS008⟩. ⟨tel-02475609⟩



Record views


Files downloads