Nilsback and Zisserman [2] noted that color and shape are the major features in flower classification. This is true only
when the flower classes considered have little intra-class variation. However, if there is a large variation within a class,
for example where species of the same type have different colors, then color may not be the best suitable feature. Hence,
in this work, we investigate the suitability of texture features in designing a system for flower classification. The flower
is segmented using a threshold-based method, and texture features, namely the color texture moments (CTMs), gray level
co-occurrence matrix (GLCM), and Gabor responses, are extracted from the segmented image and used for classification. In
considering the color texture moments, we extract moments from different color spaces of the flower images. In the gray
level co-occurrence matrix, features such as contrast, energy, entropy, correlation, and homogeneity are taken into account.
In the Gabor analysis, we have extracted the first three moments of each of the Gabor responses obtained for different scales
and orientations. These features are used for training and classification using a probabilistic neural network.