エッジ接合点，角，エッジを検出するファン・オペレータ

• Item type: SIG Technical Reports(1)
• 公開日: 1992-01-23
• タイトル: エッジ接合点，角，エッジを検出するファン・オペレータ
• タイトル / 言語: en
• タイトル: The Fan Operator for Detection of Junctions, Corners and Edges
• 言語: jpn
• 資源タイプ / 資源タイプ識別子: http://purl.org/coar/resource_type/c_18gh
• 資源タイプ: technical report
• 著者所属: 大阪大学基礎工学部
• 著者所属: 大阪大学基礎工学部
• 著者所属: 大阪大学基礎工学部
• 著者所属(英): en
• 著者所属(英): Faculty of Engineering Science
• 著者所属(英): en
• 著者所属(英): Faculty of Engineering Science
• 著者所属(英): en
• 著者所属(英): Faculty of Engineering Science
• 著者名: 中村, 高宏; 徐剛; 辻, 三郎
• 著者名(英): Takahiro, Nakamura; Gang, Xu; Saburo, Tsuji
• 論文抄録 / 内容記述タイプ: Other
• 論文抄録 / 内容記述: Both edges (1D features) and junctions and corners (2D features) are very important cues for image segmentation. This paper proposes the Fan Operator to detect junctions corners and edges directly from image. We model junctions corners and edges as intersections of straight intensity changes. The Fan Operator is a fan-shaped set of concentric circular arcs which are the first derivative of a Gaussian scaled proportionally with the radius along the axis. For each point the operator is rotated around its origin every 5 degrees and each time convolved with the image to produce a sequence of 72 values corresponding to the image changes across each direction. By analyzing these values the intensity changes are classified into a T-junction Y-junction arrow-junction a corner or a straight edge. we also show that at true junctions corners and edge points each extremum in the sequence of convolution values is also extremal among its neighborhood. By suppressing local non-extrema we eliminate those points which have directional extrema but are not true corner junction or edge points. We also show that the Fan Operator has nice properties in terms of good directional localization and signal-to-noise ratio. Finally we present experimental results on real images which show that this approach is very promising.
• 論文抄録(英) / 内容記述タイプ: Other
• 論文抄録(英) / 内容記述: Both edges (1D features) and junctions and corners (2D features) are very important cues for image segmentation. This paper proposes the Fan Operator to detect junctions, corners and edges directly from image. We model junctions, corners and edges as intersections of straight intensity changes. The Fan Operator is a fan-shaped set of concentric circular arcs which are the first derivative of a Gaussian scaled proportionally with the radius along the axis. For each point, the operator is rotated around its origin every 5 degrees, and each time convolved with the image to produce a sequence of 72 values corresponding to the image changes across each direction. By analyzing these values, the intensity changes are classified into a T-junction, Y-junction, arrow-junction, a corner, or a straight edge. we also show that at true junctions, corners and edge points each extremum in the sequence of convolution values is also extremal among its neighborhood. By suppressing local non-extrema, we eliminate those points which have directional extrema but are not true corner, junction or edge points. We also show that the Fan Operator has nice properties in terms of good directional localization and signal-to-noise ratio. Finally we present experimental results on real images, which show that this approach is very promising.
• 書誌レコードID / 収録物識別子タイプ: NCID
• 書誌レコードID / 収録物識別子: AA11131797
• 書誌情報: 情報処理学会研究報告コンピュータビジョンとイメージメディア（CVIM）; 巻 1992, 号 7(1991-CVIM-076), p. 119-126, 発行日 1992-01-23
• Notice: SIG Technical Reports are nonrefereed and hence may later appear in any journals, conferences, symposia, etc.
• 出版者 / 言語: ja
• 出版者: 情報処理学会