DefinePK

Abstract

Watershed and marker-controlled watershed transform domain methods are the one of the powerful tools for image segmentation. Segmentation and recognition are two primary stages in the development of a fully digitized flower identifier for real time use. This paper limits the following discussion to flower image segmentation only. The objective of this work is to study and explore flower detection and segmentation algorithms with watershed transform. Two variants of watershed transform with morphological gradients and wavelet coefficients are proposed. The flower segmentation problem uses watershed and marker-controlled watershed algorithm during the initial phases. This transformed into a wavelet based fusion model with binary flower images in the later stages giving reasonable segmentation outputs. The segmentation results are analysed both visually and mathematically. The average segmentation distance error (SDE) and structural similarity index (SSIM) on Oxford university flower dataset is around 0.485 and 0.786 respectively.