Clustering Algorithm Validation on Real and Synthetic Images for Promising Results in Image Segmentation and Bias Correction
Abdul Subhani Shaik, Bantupalli Premalatha, S Kirubakaran, Y. Lakshman Kumar, Ramesh Chegoni, P. Santhuja · 2024
The user’s text can be rewritten as follows: Noteworthy Arguments The task of segmenting pictures in real time can provide challenges due to their inherent homogeneity. Region-based approaches are frequently employed for image segmentation, relying on the consistency of picture intensity inside the regions of interest (ROI). However, the segmentation results generated by such algorithms sometimes exhibit inaccuracies as a result of the homogeneity of the region of interest’s intensity. This proposal presents an innovative approach for segmenting regions within an image, capable of effectively addressing variations in intensity homogeneity. Initially, the picture model that is based on intensity A local clustering criterion function is constructed to analyze the image intensities at each pixel, leading to the identification of a regional intense grouping characteristic within the picture’s homogeneities. Following the utilization of the surrounding center as a universal criterion for image dividing, the subsequent local grouping criteria are employed. This criterion establishes the concept of energy as a consequence of the level with predetermined functions that delineate a division of the area in the picture, together with a bias field that, when incorporated into an adequate level set formulation, accommodates for an enhancement in the uniformity of the picture. As a result, our proposed methodology enables the simultaneous division of the image and computation of the bias field. This is achieved by the utilization of the level set technique, which effectively reduces the energy associated with this process. In order to address the issue of uneven illumination, an approximate bias field can afterwards be employed for the purpose of additional bias correction. The efficacy of our method has been demonstrated by its use on a diverse range of actual and artificial pictures across several modalities. Notably, our method exhibits exceptional performance even in scenarios where image intensity is uniformly distributed. The experimental findings indicate that our approach exhibits superior speed, precision, and robustness against initial smooth individually models compared to the commonly employed approach. Our methodology is currently being employed for the purpose of segmenting photographs and rectifying bias, yielding highly favorable outcomes.