Image Texture Enhancement Through an Improved Grunwald-Letnikov Fractional Differential Mask

Abstract

The texture of an image is used to discriminate between segmented regions or to classify them but during image acquisition these textural features loses its contrast and gets blurred due to some natural or artificial effects. Image enhancement techniques therefore are required to emphasize and sharpen these textural features of image. The types of techniques that can sharpen texture features include point operations, where each pixel is modified according to particular equation; mask operation where each pixel is modified according to value of pixel’s neighbours. The mask operations used till date are based on integral order derivatives, these operators can enhance the high frequency information but the low frequency information like texture do not get preserved using these techniques. In order to overcome this problem concept of Fractional Differential is given. Recently, there have been lot of interest in employing Fractional Differential in various image enhancement applications like remote sensing and navigation, for segmentation or texture enhancement. Based on classical definition of Fractional Calculus many types of Fractional Differential Filter masks are developed which can improve the image texture. In this thesis, an improved Fractional Differential filter mask is designed which provides better feature enhancement. The fractional differential mask proposed in the work is derived from the definition of Grunwald-Letnikov Fractional Differential concept. The capability of this fractional differential mask is analyzed, by varying the different parameters of the fractional differential mask like intensity factor, fractional differential order and size of mask and processing the different types of images with this Fractional Differential mask and the effect of these processed images on Information Entropy and Average Gradient of image is presented. The proposed filter is then compared with other Fractional Differential filter mask by implementing other Fractional Differential masks to five images and their effects on image parameters are analyzed. It is shown that the proposed mask gives better performance in terms of Information Entropy by 0.5 than Grunwald-Letnikov and Riemann-Liouville Fractional Differential Mask.