Fig 1.
Examples of grapevine leaf shapes.
(A) Simple leaf. (B) Leaf with touching lobes [TL]. (C) Leaf with overlapping lobes [OL]. (D) Leaf with both touching and overlapping lobes [TL, OL]. Automatic threshold and contouring methods can be applied only to (A).
Fig 2.
Two-level segmentation of a grapevine leaf with overlapping lobes (OL) and touching lobes (TL).
(A) Original image obtained with a transilluminator. (B) Negative of the binay mask [T1] obtained with the maximum entropy threshold method, showing OL as small white outline. (C) Binary mask obtained with the default (bimodal) threshold method [T2], showing the whole leaf outline as a black outline. (D) Panel of nine XORed images obtained by combining slightly changed T1 and T2 threshold values (± 5% T1 and ±5% T2, respectively), showing OL as white outlines within the black outline of the leaf blade. Leaf sinuses closed by TL and OL can also be detected.
Fig 3.
Identification of overlapping and touch points.
The image shows the seven contours (labeled A to G) obtained from a composite mask of Fig 1D. [A] is the main leaf blade; [B] and [G] are OL; [C] and [F] are leaf sinuses closed by OL; [D] and [E] are leaf sinuses closed by TL. In the enlarged red frame (simplified), the green and yellow arrows show the path of the self-intersecting or slalom curve which concatenates [A], [B] and [C] contours passing through two cross points (red dots). In the enlarged blue frame, the green and yellow arrows show the path of the self-contacting curve which concatenates the [D] and [A] contours, passing through the touch point (blue dot).
Fig 4.
Concatenation of TL and OL contours.
(A) Localization of cross points (red dots) and touch points (blue dots) between the seven contours of Fig 3. (B) Result of contour concatenation. The correct link of the contours is highlighted by the sequence of black and green traces.
Fig 5.
The diagram shows the coordinate orientations (black arrows) needed to join adjacent contours (A and B) produced by OL (left panel) or TL (right panel) binary outlines. The red and blue dots indicate the cross points and touch points, respectively. The correct contour orientation and concatenation are achieved through the following steps.
Fig 6.
Visual comparison of leaf shapes obtained by considering or not considering TL or OL.
Leaves of six grapevine cultivars exhibiting TL and OL [A-F] were achieved considering (+ columns) or not considering (- columns) TL and OL. To do this, the same images were processed two times: first using the dual threshold and contouring method illustrated in this paper, and then using the default threshold method. Other cultivars that did not exhibit TL or OL [G-L] were processed using the default threshold method.
Fig 7.
Statistical comparison of leaf shapes obtained by considering or not considering TL or OL.
Principal component analysis of the first 20 EFA descriptors (elliptic harmonics) of the 90 leaf contours shown in Fig 6. The plot shows a sharp separation of the leaf shapes of A-F cultivars achieved considering (red dots) or not considering TL and OL (yellow dots). The latter are mixed with the leaves of cultivars G-L, that did not exhibit TL or OL (blue dots).
Fig 8.
Stepwise inverse EFA transform of a single leaf contour.
The image shows the leaf shape re-synthesized using the first 270 harmonics (harmonics 1 to 15, following unit steps; harmonics 60 to 270, following steps of 30). OL begin to appear from the 6th and the 12th harmonic.
Fig 9.
Separation of leaves of herbarium specimens.
(A) Original herbarium specimen. (B) Dual threshold mask. (C) Whole specimen contour. The dashed lines indicate the cutting points. (D) Single leaf contours obtained with the cut function and traced with different colors. (E) Separated leaf contours.
Fig 10.
Contours can be splitted using the cut function to remove the petiole from the leaf, to 'repair' small breaks of the leaf border and also to separate overapping leaves attached to the same twig, as shown in Fig 9. (A) Small break of the leaf border. The yellow dashed line indicates the split line. (B) The resulting 'repaired' contour.
Fig 11.
Block diagram of the ImageJ macro.
The diagram shows the main functions (shown as large blocks, identified by three-letters acronims in bold fonts) and the main sequence of operations (arrows). The complete list of operations that can be executed after the completion of each function is indicated by the 'NEXT OPTIONS' acronyms, below each block. Re-start or exit from the macro can be done at any time by menu command or by the ESC-key (not shown).