Turbopixel with fuzzy edge detection and multiregional kernel graph cuts for image segmentation

Abstract

Most existing graph cut methods does not perform well for the low-contrast and high-noisy natural images. In this paper, we present an improved image segmentation method based on TurboPixel and kernel graph cuts to solve this problem. The TurboPixel algorithm embedded with fuzzy edge detection is first applied to efficiently pre-segment the original image into homogeneous regions with precise boundary, and these regions are described as superpixels to construct the compact weighted graph. After enhancing the original image, the feature of superpixels is then represented by the Gaussian statistics in the corresponding region to replace the gray level information in the kernel graph cuts model. Additionally, in order to learn the parameters of priori knowledge accurately, we use the kernel fuzzy C-means clustering method to cluster the user interactions in this paper. Finally, considering the image local and non-local information, the energy function in the parametric kernel graph cuts is minimized to achieve the final segmentation. The experimental results demonstrate the feasibility and superiority of the proposed method in this paper.