Search

Texture plays an important role in computer vision tasks. Several methods of texture analysis are available. However, these methods are not capable of extracting rich detail in images. This paper presents a novel approach to image texture classification based on the artificial crawler model. Here, we propose a new rule of movement that moves artificial crawler agents not only toward higher intensities but also toward lower ones. This strategy is able of capturing more detail because the agents explore the peaks as well as the valleys. Thus, compared with the state-of-the-art method, this approach shows an increased discriminatory power. Experiments on the most well known benchmark demonstrate the superior performance of our approach. We also tested our approach on silk fibroin scaffold analysis, and results indicate that our method is consistent and can be applied in real-world situations.

Texture is an important visual attribute used to plant leaf identification. Although there are many methods of texture analysis, some of them specifically for interpreting leaf images is still a challenging task because of the huge pattern variation found in nature. In this paper, we investigate the leaf texture modeling based on the partial differential equations and fractal dimension theory. Here, we are first interested in decomposing the original texture image into two components f = u + v, such that u represents a cartoon component, while v represents the oscillatory component. We demonstrate how this procedure enhance the texture component on images. Our modeling uses the non-linear partial differential equation (PDE) of Perona-Malik. Based on the enhanced texture component, we estimated the fractal dimension by the Bouligand-Minkowski method due to its precision in quantifying structural properties of images. The feature vectors are then used as inputs to our classification system, based on linear discriminant analysis. We validate our approach on a benchmark with 8000 leaf samples. Experimental results indicate that the proposed approach improves average classification rates in comparison with traditional methods. The results suggest that the proposed approach can be a feasible step for plant leaf identification, as well as different real-world applications.