Abstract—In real-world images, inhomogeneity in intensities usually occurs and this makes image segmentation a difficult task. In this paper, an attempt has been made to segment the images having different level of intensity variations. The major contribution of the proposed paper is to extract accurate local intensity information using the Gaussian kernel function. This kernel function is further used to define two spatially varying fitting functions which utilize the neighborhood intensity information of each centralized pixel and guided the motion of the contour towards the desired boundary of objects in the presence of inhomogeneity. Also, multi-resolution wavelet decomposition is incorporated in the proposed method to deal efficiently with noisy images. Further, an efficient level set method by preserving the distance function has also been implemented in this paper in order to perform the comparative analysis with the proposed method. The whole framework has been tested on 15 images obtained from standard Berkeley and Weizmann databases. The results confirmed that the proposed method outperforms efficient level set algorithm method by preserving the distance function in terms of Global consistency error, Probabilistic Rand Index and Variation of Information.

Index Terms—Intensity inhomogeneity, Level set, Split-bregman, Segmentation, Wavelet

Cite: B S Saini and Saloni Pal, "LOCAL INTENSITY INFORMATION BASED IMPROVED LEVEL SET METHOD FOR IMAGE SEGMENTATION," International Journal of Electrical and Electronic Engineering & Telecommunications, Vol. 2, No. 2, pp. 58-67, April 2013.