I.J. Information Technology and Computer Science, 2017, 9, 77-84 Published Online September 2017 in MECS (http://www.mecs-press.org/) DOI: 10.5815/ijitcs.2017.09.08 Copyright © 2017 MECS I.J. Information Technology and Computer Science, 2017, 9, 77-84 Variant-Order Statistics based Model for Real- Time Plant Species Recognition Heba F. Eid Al-Azhar University, Faculty of Science, Cairo, Egypt E-mail: heba.fathy@yahoo.com Ashraf Darwish Helwan University, Faculty of Science, Cairo, Egypt E-mail: ashraf.darwish.eg@ieee.org Received: 01 June 2017; Accepted: 27 July 2017; Published: 08 September 2017 Abstract—There are an urgent need of categorizing plant by its species, to help botanist setting up a plant species database. However, plant recognition model is still very challenging task in computer vision and can be onerous and time consuming because of inefficient representation approaches. This paper, proposes a recognition model for classifying botanical species from leaf images, using combination of variant-order statistics based measures. Hence, the spatial coordinates values of gray pixels defines the qualities of texture, for the proposed model a gray-scale approach is adopted for analyzing the local patterns of leaves images using second and higher order statistical measures. While, first order statistical measures are used to extract color descriptors from leaves images. Evaluation of the proposed model shows the importance of combining variant-order statistics measures for enhancing the plant leaf recognition accuracy. Several experiments on Flavia dataset and swedish dataset are conducted. Experimental results indicates that; the proposed model yields to improve the recognition rate up to 97.1% and 94.7% for both Flavia and Swedish dataset respectively; while taking less execution time. Index Terms—Plant Recognition, Leaf Descriptors Extraction, Gray Level Co-occurrence Matrix (GLCM), Gray Level Run Length Matrices (GLRLM), leaf classification. I. INTRODUCTION Plants form an essential part of life on earth and play a major role in various areas; such as medical science and environmental science. A thorough studying and understanding of plants species is vital to increase agricultural productivity. A wide variety of plants have been named and recorded, according to statistics approximately 350,000 species of plants exist on earth. In order to study the plant effectively, plant species recognition and classification are of great importance. However, different plant species share a very close relationship to human beings. Therefore, interest for visual classification methods of plant species have grown recently [1, 2, 3]. Various plant organs have been posed such as flowers, bark, fruits or leaves for species recognition [4, 5]. Due to leaf easiness to access, carry and process; plant species recognition based on leaves has been by far the most popular methods reported in the literature [6, 7, 8, 9, 10, 11]. However, difficulty of developing leaf plant recognition models arise because leaves taxonomic have very fine differences between various species, and large variability in leaves color and texture within the same species. Also, the computer aided plant recognition procedure is very time consuming. Thus, the key issue lies in extracting leaves descriptors; which have good ability to deal with irregular textures, colors and with high intra-class variability; while taking less execution time. This motivates the design of automated leaf plant recognition model based on variant-order statistics measures. The main contribution of this paper fall in two fold: (1) the variant order- statistical approach; (2) implementation of classifying the plant species from the extracted color and texture descriptors of leaf digital image. The proposed model adopts the first order statistic measures to extract leaf color descriptors. While, second and higher order statistic measures are used to extract the textural descriptors from the digital images. Then, the combination of the extracted statistical descriptors is used for plant species recognition proposes. The main advantage of the proposed model is its simplicity and that it considers the spatial relationship, and correlation between leaf image pixels. The effectiveness of the proposed plant recognition model is evaluated by conducting several experiments on both flavia dataset and Swedish dataset. The rest of the paper is organized as follows: Section II gives the mathematical concepts of the variant order statistical measures. While, section III describes the different stages of the proposed variant-order statistics based plant recognition model. Section IV presents two leaf plant dataset; the Flavia and swedish dataset.