Publication: Extracción automática de modelos UML contenidos en imágenes
Loading...
Identifiers
Publication date
2015-11
Defense date
2016-01-19
Authors
Tutors
Journal Title
Journal ISSN
Volume Title
Publisher
Impact
Export
Abstract
Aunque parezca extraño, pese a no poder encontrar sitios web especializados en ofertar diseños de software representados mediante diagramas UML, existe una ingente cantidad de documentación a disposición de cualquiera, y que contiene dichos modelos: como imágenes en documentos textuales. Este universo de información no se encuentra fácilmente accesible para los desarrolladores porque no es posible, con la tecnología actual, buscar de forma precisa información semántica dentro de imágenes. Lo único que pueden hacer los desarrolladores es intentar buscar documentos relevantes, leerlos, y decidir si los diseños le sirven a sus intereses.
Para evitar este problema, y conseguir poner a disposición de toda la comunidad de desarrolladores centenas de miles de diseños, este trabajo pretende desarrollar la metodología necesaria para poder extraer la información textual y gráfica de las imágenes que representen diagramas UML, y convertirla en información pura UML (es decir, en modelos UML reales).
El poner a disposición de los analistas, desarrolladores de software, o interesados tal cantidad de diagramas y modelos de software permitirá la aplicación de técnicas modernas de reutilización de software basadas en la búsqueda de diagramas UML. La búsqueda de diagramas UML de todo tipo (estáticos, dinámicos, arquitecturales, de Casos de Uso etc.) mediante similitud a uno dado permitirá potenciar los desarrollos de software de calidad, controlados en el coste, y en el tiempo de desarrollo: las tres virtudes de la reutilización de software.
La complejidad de esta propuesta radica en muchos aspectos, todos ellos entrelazados: por un lado hay que considerar que la información de partida se encuentra representada con diferentes tipos de calidad, mediante bits de colores o tonos de grises. Por otro lado su semántica viene dada por la combinación de texto en lenguaje natural y estructuras gráficas. Estas estructuras gráficas tienen asociada una información semántica, accesible a la interpretación humana, que depende del tipo de diagrama. Los diagramas que representan diseños de software son documentos en formato visual con alta estructuración y contenido semántico, que se deben distinguir unos de otros. Debido a su formato en forma de imagen requieren un preprocesado mediante técnicas de visión artificial, OCR y técnicas de clusterización o clasificación basadas en aprendizaje automático. Precisamente este será el principal cometido de esta tesis: la extracción de la semántica de los diagramas en forma de imágenes encontrados en la web. La información obtenida de estos diagramas, principalmente UML, debe incluir información textual e información estructural. A la información textual se obtendrá mediante técnicas de OCR mientras que la información estructurada será detectada mediante reconocimiento de formas combinado con Inteligencia Artificial.
El resultado de esta propuesta sería una metodología que podría ser aplicada para cargar repositorios de diagramas UML a partir de imágenes existentes en internet, con vistas a su posterior aplicación y puesta a disposición de los usuarios: un GOOGLE de diagramas UML.
There are many interesting sites in the web offering reuse of source code, but no one giving the choice to identify, find and reuse design models using UML. However, even if this data seems to be sad, a simple web search can give you astonishing results: Get into GOOGLE images and search for “UML Class diagram”. Thousands of images will suddenly be available for you. The bad news: they are images. You cannot find anything on them, you cannot find them by content. You cannot compare them. You can, simple, download them. Could you be interested in working with those images, finding similar ones, etc.? In order to solve this problem, and reach hundreds of thousands of UML designs, this work intends to develop the necessary methodology to extract the textual and graphical information contented in UML based images, and convert them in, exactly, UML information (real UML Models represented in a UML object model) The possibility to offer such amount of diagrams to software analysts, software developers, or simply interested stakeholders will allow them to apply real, systematic and modern software reuse based on UML diagrams information retrieval. The possibility to find all kinds of diagrams (static, dynamic, architectural, Use Case, etc.) by similar content will strengthen software development based on the best quality, controlled cost and time to market principles: the three real benefits of Software reuse. This proposal has several difficulties in different fronts: to start with, one must consider that all the information is usually stored in low resolution images, where texts are difficult to read and understand and boxes and arrows are not properly drawn. And, on the other side, the semantics comes from the combination of text represented in Natural Language and graphical structures. These structures have associated semantic information, understandable by humans, which depend and change with the diagram types. Due to these problems, Artificial Vision, OCR, classification and automatic learning algorithm must be used in this thesis. This thesis, therefore, will attempt to extract semantic information for images representing UML Diagrams found in the web. The information extracted will be both textual and graphical. OCR technology (existing already) would be used to textual information. In order to extract graphical information a semantic model combined with AI will be used. The result of the proposal will be a methodology that will allow repositories (in the web or private) to offer UML diagrams based on (and pointing to) images found in the web, for further reuse.
There are many interesting sites in the web offering reuse of source code, but no one giving the choice to identify, find and reuse design models using UML. However, even if this data seems to be sad, a simple web search can give you astonishing results: Get into GOOGLE images and search for “UML Class diagram”. Thousands of images will suddenly be available for you. The bad news: they are images. You cannot find anything on them, you cannot find them by content. You cannot compare them. You can, simple, download them. Could you be interested in working with those images, finding similar ones, etc.? In order to solve this problem, and reach hundreds of thousands of UML designs, this work intends to develop the necessary methodology to extract the textual and graphical information contented in UML based images, and convert them in, exactly, UML information (real UML Models represented in a UML object model) The possibility to offer such amount of diagrams to software analysts, software developers, or simply interested stakeholders will allow them to apply real, systematic and modern software reuse based on UML diagrams information retrieval. The possibility to find all kinds of diagrams (static, dynamic, architectural, Use Case, etc.) by similar content will strengthen software development based on the best quality, controlled cost and time to market principles: the three real benefits of Software reuse. This proposal has several difficulties in different fronts: to start with, one must consider that all the information is usually stored in low resolution images, where texts are difficult to read and understand and boxes and arrows are not properly drawn. And, on the other side, the semantics comes from the combination of text represented in Natural Language and graphical structures. These structures have associated semantic information, understandable by humans, which depend and change with the diagram types. Due to these problems, Artificial Vision, OCR, classification and automatic learning algorithm must be used in this thesis. This thesis, therefore, will attempt to extract semantic information for images representing UML Diagrams found in the web. The information extracted will be both textual and graphical. OCR technology (existing already) would be used to textual information. In order to extract graphical information a semantic model combined with AI will be used. The result of the proposal will be a methodology that will allow repositories (in the web or private) to offer UML diagrams based on (and pointing to) images found in the web, for further reuse.
Description
Keywords
Diseño de software, UML, Proceso de imágenes, Reconocimiento de formas, Recuperación de la información, Inteligencia artificial