Superpixel Segmentation using Symmetric Kernel Function and it’s Application in Image Forgery Detection

Superpixel segmentation plays an important role in the area of computer vision. In this paper, we present a symmetric kernel function based algorithm to produce Superpixel segmentation of an image. The proposed method adopts $L_{1}$ norm for similarity measure in objective function of normalized cuts. The weighted k-means clustering function is minimized by the kernel function property defined over the weight and mapping function in optimized objective function. The positivity condition achieved in the weighted function and mapping function has formed a way for the application of kernel trick to obtain clusters. It is observed that the optimum points of both normalized cuts and weighted k-means are same. The proposed method achieves reduced computational complexity as it adopts kernel function for computation of weights and mapping function required for minimizing objective function, where as the traditional methods which compute Eigen values results in high complexity. The experimental results show that the boundary adherence and compactness of the proposed method is better when compared with the existing techniques. The proposed method is further applied to detect copy move forgery in digital images for performance evaluation.