Workshop on Shape, Contour and Grouping

形状、轮廓和分组研讨会

• 批准号: 9712426
• 负责人: David Forsyth
• 金额: $ 1万
• 依托单位: University of California-Berkeley
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 1997
• 资助国家: 美国
• 起止时间: 1997-07-01 至 1999-06-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=9712426&HistoricalAwards=false
• 关键词: Workshop; Shape; Contour; Grouping

A new application for object recognition, indexing image sets in digital libraries, demands recognition strategies that are (1) effective in cluttered scenes and (2) effective for objects recognised at a class level - for example, people or animals. These needs can be met by recognition algorithms that interpret the world in terms of stylised properties in shape, shading, texture and colour, which are caled 'primitives.' In this approach, objects are represented by a collection of spatial relationships between primitives. The primitives themselves are used as a source of grouping constraints. Recognition proceeds by grouping together features or regions that could represent a primitive, and then assembling groups of primitives that could have come from an object. There is no explicit 3D reasoning, but the grouping processes may contain a great deal of implicit 3D reasoning. This approach to shape and to recognition unites themes studied in active research programs in computer vision, both in the United States and in the European Community. A joint workshop, involving both US and E.C. participants (for whom E.C. funding will be available), will explore representations that use grouping and symmetry as a recognition strategy. Topics will include: shading primitives, image shape constraints that result from volumetric primitives, building grouping-based representations and overall strategies for object recognition. The workshop will be held in Erice, Sicily, in early 1998. The proceedings will be published as a volume in the Springer Lecture Notes in Computer Science series.

对象识别的新应用，即在数字图书馆中索引图像集，需要以下识别策略：(1) 在杂乱的场景中有效；(2) 在类级别识别对象（例如人或动物）有效。 这些需求可以通过识别算法来满足，这些算法根据形状、阴影、纹理和颜色等风格化属性（称为“基元”）来解释世界。 在这种方法中，对象由基元之间的空间关系的集合来表示。 原语本身用作分组约束的来源。 识别过程是通过将可以表示基元的特征或区域分组在一起，然后组装可能来自对象的基元组来进行的。 没有显式的 3D 推理，但分组过程可能包含大量隐式的 3D 推理。 这种形状和识别方法结合了美国和欧洲共同体计算机视觉活跃研究项目中研究的主题。 由美国和欧洲参与者（将获得欧洲委员会资助）参加的联合研讨会将探讨使用分组和对称作为识别策略的表示。 主题将包括：着色基元、由体积基元产生的图像形状约束、构建基于分组的表示以及对象识别的总体策略。 该研讨会将于 1998 年初在西西里岛埃里切举行。会议记录将作为 Springer 计算机科学讲义系列中的一卷出版。