An efficient fractal measure for image texture recognition

Fractal measures like fractal dimension (FD), lacunarity, succolarity measure the geometrical complexity of objects and could be used to describe texture information of the images. For this purpose different box counting algorithms were developed to estimate FD. However the existing box-counting methods usually suffer from under counting or over counting, introducing difficulties in obtaining the exact value of the FD. This paper focuses on the box-counting's power in uniquely identifying patterns and presents a new approach which considers the aggregate effects of all the gray levels in the boxes, rather than considering only two gray levels, (min and max) as in the case of traditional differential box-counting method. The proposed method uses new counting measure based on volume percentage of the gray levels inside the boxes. Results from experiments tabulated to depict the improved effect of the proposed method in recognition of the noisy test images from Brodatz Texture and normal test images from CASIA-V3 Iris Databases.