Financial support

By my self

Multi-scale analisys of leaf textures applied in the plant classification


Mauricio Falvo, Odemir M. Bruno


Much has been researched with respect to the characterization and identification of plant species through of biometric data. Some methods are based in morphometric measures of leaf venation system, others in internal venation system[1]. These methods were able to establish relationships among different species which were confirmed by comparison of genetic markers[2]. However, for some species of plants only the shape attribute or venation system are not enough for a reliable classification. Thus, recent studies have been focused efforts for characterization and classification of plant species, through analysis of the texture of the leaf surface, both the macroscopic level[3] as microscopic [4].

What can be noted is that these jobs rarely consider jointly the different levels of scale observation, that is, did not there a multi-scale approach [5]. So, the aim of this project is the research, implementation, development and test of methods for analyzing multi-scale applied in vegetal tissue textures, in order to classify them. Initially the adaxial leaf surface (front face of the leaf) and abaxial (back side of the leaf) will be studied. However, other parts of the plant can be investigated, such as petals or plant tissue sections (images obtained by microscopy). The multi-scale approach will be done with DWT, Gabor transform and Laplacian pyramid, combining them with statistical methods of feature extraction and others characterization methods such as fractal dimension [2] and complex networks.

Examples of real multi-scale texture images of abaxial surface of Passiflora elegans N. ; 100x (a,b); 200x(c) and 300x (d)