Flower Recognition System based on Color and GIST Features

Recognition of flower species from natural images is a challenging task as it involves a significant amount of preprocessing to separate the actual flower from its surrounding background. The next big challenge is to represent these images mathematically so that a classification algorithm can be put into place to classify them. In this paper, the task of segmenting flower images from their natural background is achieved by proposing and implementing a segmentation method based on average color and the variance of the color distribution. Flower images are encoded into mathematical feature vectors using combined color and GIST features. Tests involving the aforementioned features individually along with several other image features, also tested individually and in combination, show that the proposed representation works best for the task. Finally, a classification model, based on Support Vector Machine (SVM), is trained. The trained model can distinguish between 12 different classes of flowers. The proposed approach displays an accuracy of 85.93%.