An Improved Grunwald-Letnikov Fractional Differential Mask for Video Enhancement

Abstract

The texture of images and videos is utilized to segregate or classify between segmented regions. But at the time of image and video acquisition, textural features lose its contrast. During image/video processing, these textural features get blurred due to some natural or artificial effects. Therefore, image/video enhancement techniques are required to emphasize and sharpen these textural features. Generally, the problem of oversaturation arises in the conventional enhancement methods. However, the fact is that the conventional methods enhance the images/video frames, but at the same time introduces significant unnatural noises in images/video frames. Moreover, the brightness of dark region in images/frames increases. In the proposed enhancement approach, improved G-L method is applied on the images for benchmarking. In case of videos, first of all frames are excerpted from input video and then, non-linear mask which is created using improved G-L method is applied on the video frames. For enhancing the textural information and to achieve recognition accuracies in images/video frames, a fractional differential operator is realized (two dimensional) which is an improved version of Grunwald-Letnikov (G-L) based differential operator. Lagrange’s method of 3-point interpolation is applied to simple G-L equation for creation of mask. Experimental results demonstrate that, the proposed mathematical fractional differential method efficiently enhances the images/video frames without the oversaturation problem. Moreover, this approach provides high degree of enhancement for low light and bright light images/videos. Enhancement method is applied on various videos and it demonstrates that the presented method enhances the quality of frames and apprizes the accuracy. The enhancement approach is compared to standard and existing enhancement approaches and the results show that this model outperforms the existing models. To figure out the enhancement criteria, average gradient, peak signal to noise ratio (PSNR), structural similarity index and information entropy values are calculated. It is also shown that the proposed mask gives better performance in terms of Information Entropy and Average Gradient than Grunwald-Letnikov and Riemann-Liouville Fractional Differential Masks.