Image segmentation is a fundamental process in most image processing, pattern recognition, and computer vision applications. It separates the desired objects from the set of desired and undesired objects in an image. This paper presents such an image segmentation method.
The author presents an image segmentation model described in three terms: regularization, data fidelity, and decomposing an image into two parts, piecewise smooth and oscillating texture. This model “integrates image decomposition and image segmentation into a unified model within [a] fuzzy region competition framework.” He further divides the minimization problem into four subproblems: dictionary learning and structure-texture decomposition, optimization of the piecewise constant image, membership functions, and auxiliary functions.
The author tested the proposed method with total variation regularization fuzzy region competition, wavelet regularization, a graph-cut-based method using principal component analysis, and Gabor-based methods on both noiseless images and noisy images with Gauss white noise (GWn). For experiments, he used synthetic and real-world natural images with various resolutions and a complex texture image.
According to the author, his algorithm “can obtain cartoon and texture features directly during segmentation,” and can perform image segmentation, image decomposition, and image denoising. This paper is worth reading for those working in this area.
The author claims that, “in terms of segmentation results, iterations, central processing unit (CPU) time, and peak signal-to-noise ratio (PSNR), the experiments show the applicability and effectiveness of the proposed method.” Further, the solution to the minimization problem in the appendix makes this paper an interesting read.