Content-based image retrieval system for plant leaf database using texture

• Main Author: Hussein, Ahmed Naser
• Format: Thesis
• Language: English
• Published: 2011
• Online Access: http://psasir.upm.edu.my/id/eprint/41795/1/FK%202011%2020R.pdf; http://psasir.upm.edu.my/id/eprint/41795/

Automatic plant leaf images retrieval system help the students and the researchers in botany field. It does so by overcoming limitations associated by the system such as the domain knowledge requirement and the time consumption. It also helps in learning process where the retrieval will speed up the search of any plant species and gives better experience to the students to familiarize themselves with the plant species. The motivation of this work was driven by inherent difficulties of the manual classification of plant leaf images. To achieve that, an automatic, fast, and robust content based image retrieval (CBIR) system is designed. The richness and uniqueness of plant leaf texture is used in this work as a principal feature in classifying the plant leaf species. A study on the texture extraction approach for plant leaf image is crucial in designing an effective image retrieval system.To classify plant leaf image, CBIR system is employed which extract the leaf texture and then use the extracted feature to compare against the gallery for similarity measurement. The texture extraction is accomplished using Discrete Wavelet Transformation (DWT) incorporating with entropy measurement which enhances the classification of images. The dataset for this experimental work has been obtained from the American National Herbarium Collections. The dataset offers wide diversity of rotation, noise, luminance and scale on plant leaf image. The experiments have been performed on seven plant species that consist of 280 images. To evaluate the robustness of the system, the experiments are repeated on 92 species that consist of 3597 images. The proposed framework yields a correct classification rate of 92.5% , 85.92% of average precision rate for top 5 images, 71.9% of average recall rate for top 100 images, length of feature vector is 36, and the average retrieval time is 1.0656 seconds only on overall system framework. The results were compared with another CBIR system which is based on Gray Level Co-occurrence Matrix (GLCM) and then showed better performance in terms of evaluation of images classification and retrieval.