Loading...
Publication
Performance assessment of a B-COSFIRE algorithm extension for craquelure segmentation in artworks
| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 8.98 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
As obras de arte, particularmente as pinturas, representam peças culturalmente valiosas com história e domínio técnico. Com o passar do tempo, estas peças desenvolvem fissuras, que são redes finas e intrincadas nas superfícies. Estes padrões proporcionam informação relevante acerca da antiguidade, origem e estado das obras de arte, contribuindo assim para a conservação e legitimação das mesmas.
Neste contexto, este projeto aborda o complexo desafio da segmentação de padrões de fissuras na análise de obras de arte, adaptando técnicas avançadas de processamento de imagem de outros campos. Mais concretamente, é apresenta uma versão modificada do algoritmo B-COSFIRE, originalmente concebido para a deteção de vasos sanguíneos na retina, adaptado e otimizado para segmentar com precisão fissuras em obras de arte diversificadas. Esta adaptação tem como objetivo ultrapassar as limitações dos métodos existentes e disponibilizar a precisão necessária para a conservação e análise eficazes da arte.
De forma a validar o algoritmo, foram criados dois conjuntos de dados especializados, consistindo num conjunto de dados em escala de cinzentos da coleção Bucklow, representando vários períodos históricos e estilos artísticos, e um conjunto de dados RGB de pinturas de domínio público, capturando a complexidade das obras de arte coloridas. Os resultados mostraram um melhor desempenho do algoritmo B-COSFIRE modificado na segmentação de padrões de fissuras em obras de arte em escala de cinzentos, enquanto obras de arte de cores mais complexas introduzem padrões de segmentação mais desafiantes, mantendo, no entanto, um melhor desempenho em comparação com o estado da arte.
Assim, este projeto apresenta uma solução para colmatar a ausência de técnicas avançadas de processamento de imagem e conservação de arte, oferecendo uma abordagem inovadora para a análise de fissuras. Dessa forma, ao demonstrar como o algoritmo B-COSFIRE modificado pode ser eficazmente adaptado para este propósito, também se destaca a importância de integrar tecnologias de ponta na preservação do patrimônio cultural. Além disso, o trabalho futuro inclui a extensão dos conjuntos de dados, a exploração de abordagens de aprendizagem profunda, a melhoria das técnicas de pré-processamento e pós-processamento, e a conceção de um filtro independente do domínio para expandir a aplicabilidade do algoritmo para além da conservação de obras de arte.
Works of art, particularly paintings, represent culturally valuable pieces with history and technical mastery. Over time, these pieces develop cracks, which are fine, intricate networks on their surfaces. These patterns provide relevant information about the antiquity, origin and condition of works of art, thus contributing to their conservation and legitimization. In this context, this project addresses the complex challenge of segmenting crack patterns in the analysis of works of art, adapting advanced image processing techniques from other fields. More specifically, it presents a modified version of the B-COSFIRE algorithm, originally designed for the detection of blood vessels in the retina, adapted and optimized to accurately segment cracks in diverse works of art. This adaptation aims to overcome the limitations of existing methods and provide the precision required for effective art conservation and analysis. In order to validate the algorithm, two specialized datasets were created, consisting of a greyscale dataset from the Bucklow collection, representing various historical periods and artistic styles, and an RGB dataset of public domain paintings, capturing the complexity of colored artworks. The results showed a better performance of the modified B-COSFIRE algorithm in segmenting crack patterns in greyscale artworks, while more complex colored artworks introduce more challenging segmentation patterns, while still maintaining a better performance compared to the state of the art. Thus, this project presents a solution to address the lack of advanced image processing and art conservation techniques by offering an innovative approach to crack analysis. Thus, by demonstrating how the modified B-COSFIRE algorithm can be effectively adapted for this purpose, the importance of integrating cutting-edge technologies in the preservation of cultural heritage is also highlighted. In addition, future work includes extending the datasets, exploring deep learning approaches, improving pre-processing and post-processing techniques, and designing a domain-independent filter to expand the applicability of the algorithm beyond the conservation of works of art.
Works of art, particularly paintings, represent culturally valuable pieces with history and technical mastery. Over time, these pieces develop cracks, which are fine, intricate networks on their surfaces. These patterns provide relevant information about the antiquity, origin and condition of works of art, thus contributing to their conservation and legitimization. In this context, this project addresses the complex challenge of segmenting crack patterns in the analysis of works of art, adapting advanced image processing techniques from other fields. More specifically, it presents a modified version of the B-COSFIRE algorithm, originally designed for the detection of blood vessels in the retina, adapted and optimized to accurately segment cracks in diverse works of art. This adaptation aims to overcome the limitations of existing methods and provide the precision required for effective art conservation and analysis. In order to validate the algorithm, two specialized datasets were created, consisting of a greyscale dataset from the Bucklow collection, representing various historical periods and artistic styles, and an RGB dataset of public domain paintings, capturing the complexity of colored artworks. The results showed a better performance of the modified B-COSFIRE algorithm in segmenting crack patterns in greyscale artworks, while more complex colored artworks introduce more challenging segmentation patterns, while still maintaining a better performance compared to the state of the art. Thus, this project presents a solution to address the lack of advanced image processing and art conservation techniques by offering an innovative approach to crack analysis. Thus, by demonstrating how the modified B-COSFIRE algorithm can be effectively adapted for this purpose, the importance of integrating cutting-edge technologies in the preservation of cultural heritage is also highlighted. In addition, future work includes extending the datasets, exploring deep learning approaches, improving pre-processing and post-processing techniques, and designing a domain-independent filter to expand the applicability of the algorithm beyond the conservation of works of art.
Description
Keywords
Segmentação Supervisionada de Estruturas B-COSFIRE Processamento de imagem Estruturas curvilíneas