Histogram-based fast and robust image clustering using stochastic fractal search and morphological reconstruction

Partitional clustering-based image segmentation is one of the most significant approaches. K-means is the conventional clustering techniques even though very sensitive to noise and easy convergences to local optima depending on the initial cluster centers. In addition, the computational time of K-means algorithm is also very high due to the repetitive distance calculation between pixels and cluster centers. In order to solve these problems, this paper presents a Histogram-based Fast and Robust Crisp Image Clustering (HFRCIC) technique. Local spatial information is often introduced to an objective function to improve the noise robustness of the clustering technique. At first, the local spatial information has been introduced into HFRCIC by incorporating morphological reconstruction which assures noise-immunity as well as image detail-preservation. Then clustering has been executed depending on gray levels in the place of pixels of the image. As result, the execution time is low as the number of gray levels is usually much smaller than the number of pixels in the image. Due to the random initialization of centers, HFRCIC easily stuck into local optima as HFRCIC is greedy in nature and an efficient local optimizer. Therefore, Nature-Inspired Optimization Algorithms (NIOA) are successfully employed to overcome the problem within reasonable computational time. Here, Stochastic Fractal Search (SFS) has been employed to find the optimal cluster centers. The experimental study has been performed over synthetic images, real-world images and white the gray level conversion of RGB imaged for white blood cell (WBC) segmentation. Visual and numerical results indicate the superiority of the proposed HFRCIC with SFS(HFRCIC-SFS) over state-of-the-art image segmentation algorithms and NIOA-based crisp image clustering techniques.