An Adaptive Shape Recognition System for Leaves

Abstract

Shape recognition is a challenging problem within the study of image processing and image understanding. The thesis presents a study of an adaptive shape recognition system based on image segmentation, shape description and,neural networks. An adaptive method was chosen so that it could adapt to the shapes of interest to achieve the broad application of the shape recognition system. An image segmentation technique was implemented to automatically separate the objects in the image. A shape description technique was used to extract characteristics of a shape through calculation of its Projection Length Sequences (PLS) at different angles. Algorithms were developed that could tolerate a certain amount of occlusion and noise on the shape. A neural network was designed and trained as an adaptive classifier to recognize the PLS of different shapes. The training algorithm helps the system to recognize shapes with a variety of orientations. The experimental results presented in this thesis show that the system gives a 93% average recognition rate for a set of images that include tree leaves with three different shapes.