S. Lengauer, Peter Houska, R. Preiner, E. Trinkl, S. Karl, I. Sipiran, Benjamin Bustos, T. Schreck
{"title":"基于深度学习的彩陶几何纹饰交互标注","authors":"S. Lengauer, Peter Houska, R. Preiner, E. Trinkl, S. Karl, I. Sipiran, Benjamin Bustos, T. Schreck","doi":"10.1515/itit-2022-0007","DOIUrl":null,"url":null,"abstract":"Abstract In Greek art, the phase from 900 to 700 BCE is referred to as the Geometric period due to the characteristically simple geometry-like ornamentations appearing on painted pottery surfaces during this era. Distinctive geometric patterns are typical for specific periods, regions, workshops as well as painters and are an important cue for archaeological tasks, such as dating and attribution. To date, these analyses are mostly conducted with the support of information technology. The primitives of an artefact’s ornamentation can be generally classified into a set of distinguishable pattern classes, which also appear in a similar fashion on other objects. Although a taxonomy of known pattern classes is given in subject-specific publications, the automatic detection and classification of surface patterns from object depictions poses a non-trivial challenge. Our long-term goal is to provide this classification functionality using a specifically designed and trained neural network. This, however, requires a large amount of labelled training data, which at this point does not exist for this domain context. In this work, we propose an effective annotation system, which allows a domain expert to interactively segment and label parts of digitized vessel surfaces. These user inputs are constantly fed back to a Convolutional Neural Network (CNN), enabling the prediction of pattern classes for a given surface area with ever increasing precision. Our work paves the way for a fully automatic classification and analysis of large surface pattern collections, which, with the help of suitable visual analysis techniques, can answer research questions like pattern variability or change over time. While the capability of our proposed annotation pipeline is demonstrated at the example of two characteristic Greek pottery artefacts from the Geometric period, the proposed methods can be readily adopted for the patternation in any other chronological periods as well as for stamped motifs.","PeriodicalId":43953,"journal":{"name":"IT-Information Technology","volume":null,"pages":null},"PeriodicalIF":1.0000,"publicationDate":"2022-08-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Interactive annotation of geometric ornamentation on painted pottery assisted by deep learning\",\"authors\":\"S. Lengauer, Peter Houska, R. Preiner, E. Trinkl, S. Karl, I. Sipiran, Benjamin Bustos, T. Schreck\",\"doi\":\"10.1515/itit-2022-0007\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"Abstract In Greek art, the phase from 900 to 700 BCE is referred to as the Geometric period due to the characteristically simple geometry-like ornamentations appearing on painted pottery surfaces during this era. Distinctive geometric patterns are typical for specific periods, regions, workshops as well as painters and are an important cue for archaeological tasks, such as dating and attribution. To date, these analyses are mostly conducted with the support of information technology. The primitives of an artefact’s ornamentation can be generally classified into a set of distinguishable pattern classes, which also appear in a similar fashion on other objects. Although a taxonomy of known pattern classes is given in subject-specific publications, the automatic detection and classification of surface patterns from object depictions poses a non-trivial challenge. Our long-term goal is to provide this classification functionality using a specifically designed and trained neural network. This, however, requires a large amount of labelled training data, which at this point does not exist for this domain context. In this work, we propose an effective annotation system, which allows a domain expert to interactively segment and label parts of digitized vessel surfaces. These user inputs are constantly fed back to a Convolutional Neural Network (CNN), enabling the prediction of pattern classes for a given surface area with ever increasing precision. Our work paves the way for a fully automatic classification and analysis of large surface pattern collections, which, with the help of suitable visual analysis techniques, can answer research questions like pattern variability or change over time. While the capability of our proposed annotation pipeline is demonstrated at the example of two characteristic Greek pottery artefacts from the Geometric period, the proposed methods can be readily adopted for the patternation in any other chronological periods as well as for stamped motifs.\",\"PeriodicalId\":43953,\"journal\":{\"name\":\"IT-Information Technology\",\"volume\":null,\"pages\":null},\"PeriodicalIF\":1.0000,\"publicationDate\":\"2022-08-31\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"IT-Information Technology\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/10.1515/itit-2022-0007\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"Q4\",\"JCRName\":\"COMPUTER SCIENCE, INFORMATION SYSTEMS\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"IT-Information Technology","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1515/itit-2022-0007","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q4","JCRName":"COMPUTER SCIENCE, INFORMATION SYSTEMS","Score":null,"Total":0}
Interactive annotation of geometric ornamentation on painted pottery assisted by deep learning
Abstract In Greek art, the phase from 900 to 700 BCE is referred to as the Geometric period due to the characteristically simple geometry-like ornamentations appearing on painted pottery surfaces during this era. Distinctive geometric patterns are typical for specific periods, regions, workshops as well as painters and are an important cue for archaeological tasks, such as dating and attribution. To date, these analyses are mostly conducted with the support of information technology. The primitives of an artefact’s ornamentation can be generally classified into a set of distinguishable pattern classes, which also appear in a similar fashion on other objects. Although a taxonomy of known pattern classes is given in subject-specific publications, the automatic detection and classification of surface patterns from object depictions poses a non-trivial challenge. Our long-term goal is to provide this classification functionality using a specifically designed and trained neural network. This, however, requires a large amount of labelled training data, which at this point does not exist for this domain context. In this work, we propose an effective annotation system, which allows a domain expert to interactively segment and label parts of digitized vessel surfaces. These user inputs are constantly fed back to a Convolutional Neural Network (CNN), enabling the prediction of pattern classes for a given surface area with ever increasing precision. Our work paves the way for a fully automatic classification and analysis of large surface pattern collections, which, with the help of suitable visual analysis techniques, can answer research questions like pattern variability or change over time. While the capability of our proposed annotation pipeline is demonstrated at the example of two characteristic Greek pottery artefacts from the Geometric period, the proposed methods can be readily adopted for the patternation in any other chronological periods as well as for stamped motifs.