Rough and accurate segmentation of natural color images using fuzzy region-growing algorithm

Abstract

We present a rough and an accurate segmentation of nat ural color images using a fuzzy region-growing algorithm.In the proposed met h od, the color diff erence and the color gradient are used as t he pixel features to produce an accu rate segmentation, while t h e localfractal dimension is used as the region feature to yield a rough segmentation in a natu ral color image. The effectiveness of the proposed method is confirmed through computer simulations that demonstrate a rough segmentation at the fine-texture regions and an accu rate segmentation at the strong-edge regions simultaneously.

Region-based fit of color homogeneity measures for fuzzy image segmentation

“…Based on the idea of fuzzy topology, introduced by Rosenfeld [24], and the use of fuzzy connectivity to measure the relationship between any pair of pixels, these techniques obtain the fuzzy region by calculating the connectivity with respect to the seed of a region [1,7,13,17,21,25].…”

Fuzzy Based Image Segmentation

“…This scheme minimizes the fuzziness in the thresholded description and allows for accommodating the variations in the gray values within each region. Examples of popular fuzzy clustering approaches are the fuzzy c-mean (FCM) [11] and the fuzzy region merging [12][13][14]. For example, a fuzzy region-growing algorithm was used to segment natural color images in [14].…”

“…Examples of popular fuzzy clustering approaches are the fuzzy c-mean (FCM) [11] and the fuzzy region merging [12][13][14]. For example, a fuzzy region-growing algorithm was used to segment natural color images in [14]. In [15], the thresholds associated with membership functions were computed by means of fuzzy c-mean clustering (FCM) to perform fuzzy segmentation.…”

On Significant Crisp Representatives of Fuzzy Regions in Colour Images

“…In addition, crisp techniques [14], [7] are not well suited to solve the problem of obtaining the model of a single region, in which a single seed is specified, since at least two seeds are needed, and the model obtained depends not only on the seed indicating the region of interest, but of the number and position of the rest of the seeds. Most of the fuzzy extensions of region growing techniques have the same problem [1], [15], [12].…”