Color Image Segmentation using Fast Fuzzy C-Means Algorithm

Abstract

This paper proposes modified FCM (Fuzzy C-Means) approach to colour image segmentation using JND (Just Noticeable Difference) histogram. Histogram of the given colour image is computed using JND colour model. This samples the colour space so that just enough number of histogram bins are obtained without compromising the visual image content. The number of histogram bins are further reduced using agglomeration. This agglomerated histogram yields the estimation of number of clusters, cluster seeds and the initial fuzzy partition for FCM algorithm. This is a novell approach to estimate the input parameters for FCM algorithm. The proposed fast FCM(FFCM) algorithm works on histogram bins as data elements instead of individual pixels. This significantly reduces the time complexity of FCM algorithm. To verify the effectiveness of the proposed image segmentation approach, its performance is evaluated on Berkeley Segmentation Database(BSD). Two significant criteria namely PSNR(Peak Signal to Noise Ratio) and PRI (Probabilistic Rand Index) are used to evaluate the performance. Although results show that the proposed algorithm applied to the JND histogram bins converges much faster and also gives better results than conventional FCM algorithm, in terms of PSNR and PRI.Key Words: Colour Image Segmentation, JND Histogram, Fuzzy C-means Clustering, Fast FCM IntroductionSegmentation involves partitioning an image into a set of homogeneous and meaningful regions, such that the pixels in each partitioned region posses an identical set of properties. Fuzzy set theory has been extensively applied in the area of image segmentation. The concepts of fuzzy first order and second order statistic like fuzzy histogram and co-occurrence matrices have been presented in [1,2]. These measures have been demonstrated to yield excellent segmentation results for the images having bimodal or multimodal histograms. The fuzzy second order statistics characterize textured images in a better way than their corresponding hard counterparts [3]. A number of research reports indicate the superiority of fuzzy sets in segmenting an image over its crisp counterparts.Fuzzy approaches to pixel classification have found applications in problems where precise knowledge about the pattern classes is not available, large number of pattern samples are not available for statistical estimation of parameters and patterns have partial membership to different classes. In many real world images, the clusters are not disjoint and a simple pattern or a pixel in an image may belong to different clusters. For example, a particular pixel in the sky may belong partly to sky and partly to the cloudy-sky class. Fuzzy approaches to supervised pattern classification and clustering may be found in [4]. Clustering methods are considered as an unsupervised classification technique, where there is no need for prior

Remote Photoplethysmography (rPPG): A State-of-the-Art Review

“…In the case, where only the mouth area is focused to obtain HR, it can impact the error rate when a person is talking or laughing as teeth do not provide a pulse signal to obtain HR. To cater to this problem, researchers [100] introduced neural networks for skin classification [163] to determine skin colour from the face image. This ROI (skin pixel) was labelled and tracked using a mean-shift tracker, which iteratively shifts points of data to the average of points of data nearby [108].…”

An evaluation on changes in hippocampus size for cognitively normal, mild cognitive impairment, and Alzheimer’s disease patients using fuzzy membership function

“…For accurate segmentation of a region in the brain, it is necessary to strip the skull part from the MR images [28]. After comparing the performance of five popular segmentation techniques namely Region-growing [29], Region splitting & Merging [30], K -Means algorithm [31], Histogram-based algorithm [32], and Fuzzy C Means [33] for 50 MRI images, it is found that the Histogram Based Thresholding technique gives the highest accuracy among all these algorithms [34]. Hence, for skull stripping, we have used the Histogram Based Thresholding technique.…”

An Overview of Image Segmentation Algorithms

“…Bhoyar and Kakde [34] proposed modified FCM (Fuzzy C-Means) approach to color image segmentation using JND (Just Noticeable Difference) histogram. The approach is compared with the (fuzzy c-means) FCM for segmentation, which is faster than FCM.…”