Differential box-counting (DBC) method has been widely used to calculate fractal dimension. analyzing the result of fractal dimension with background image by estimating this traditional method,we find that this algorithm has the limitation to estimate fractal dimension of images with background accurately. In order to solve this issue ,in this paper, an improved differential box-counting(IDBC) method has been proposed to eliminate influence of background on fractal dimension .The mainly improved steps of IDBC as follow: firstly ,the background pixels values G<sub>b</sub> of image need to be found out using probability .Secondly, the n<sub>q</sub> ( n<sub>q</sub> is the numbers of boxes in a grid) is set to zero when the maximum and minimum in a grid are equal to image background values. To validate IDBC method’s performance ,we designed an experiment that the fractal dimension of both original texture images and these ones put in different size frame with black-background are estimated and compared through four different algorithms, including DBC, relevant differential box-counting (RDBC) method, shifting differential box-counting (SDBC) method and IDBC method. The experimental results demonstrate that the IDBC method developed in this work has the ability to improve the measurement accuracy by avoiding the influence caused by background.
Fractal dimension is an important quantitative characteristic of a image, which can be widely used in image analysis. Differential box-counting method which is one of many calculation methods of a fractal dimension has been frequently used due to its simple calculation . In differential box-counting method, a window size M is limited in the integer power of 2. It leads to inaccurate calculation results of a fractal dimension. Aiming at solving the issues , in this paper, an improved algorithm is discussed that the window size M has been improved to be able to accommodate non-integer power of 2, and making the calculated fractal dimension error smaller. In order to verify superiority of the improved algorithm, the values of fractal dimension are regarded as parameters, and are applied for image segmentation combined with Ostu algorithm . Both traditional and improved differential box-counting methods are respectively used to estimate fractal dimensions and do threshold segmentation for a thread image . The experimental results show that image segmentation details by improved differential box-counting method are more obvious than that by traditional differential box-counting method, with less impurities, clearer target outline and better segmentation effect.