In the present study, we quantitatively evaluated the shape of the tip part of petal in sacred lotus using P-type Fourier descriptors (PFDs), and assessed the validity of this method through statistical analyses of PFDs. From fragmented lotus petals, which were obtained by cutting each bowl-shaped petal into four fragments, we extracted the contour of the tip part of petal and delineated the contour shape using PFDs up to the 8
th order. The first principal component of PFDs accounted for 50% of the total variation, and the cumulative contribution of the first five components reached 80%. The shape variation explained by each principal component could be visualized by inverse Fourier transformation. The 1
st and 2
nd components provided good measures of the sharpness and asymmetry of the tip part of petal, respectively. The 3
rd and 4
th components explained the features of the apex of the tip part. We assessed shape variations within a flower based on petals of two varieties, and found that there was a systematic variation in the 1st principal component. As a result of ANOVA, the varietal effect was significant at the 0.1% level in the 1
st and 3
rd principal components, indicating that the shape characteristics accounted for by these components reflected the among-variety variation well. The varietal mean of these two components showed a continuous distribution, indicating that it is difficult to grade these shape characteristics. Finally, we conducted a variety classification based on principal component scores of PFDs using a support vector machine. The rate of correct classification of 7 varieties was estimated to be as high as 85% based on leave-one-out cross-validation. Our results indicated that principal component analysis (PCA) of PFDs extracted the independent shape characteristics of petal of sacred lotus and that these characteristics could be efficiently used in the classification of sacred lotus varieties. Since PFDs can be applied to the description of an open curve, it is considered that the method based on PFDs could widen the range of application of shape analysis based on image processing, and could become one of the effective evaluation techniques for crop organ shapes. This report is the first in which PFDs were used and found to be applicable to the shape described with an open curve, in the quantitative evaluation of plant shape.
View full abstract