Robust Blob Detection and Refinement via Iterative Anisotropic Gaussian Filtering and Clustering

Blob detection is a crucial task in medical image analysis, particularly for identifying and localizing regions of interest. This paper addresses the challenges of blob detection in medical images, including the detection of diverse blob shapes (both circular and elliptical), robustness against affine transformations and noise, and the separation of overlapping blobs. To this end, we propose a novel two-stage blob detection method. In the first stage, an iterative approach is employed to generate anisotropic blob representations, effectively suppressing non-blob regions and overlapping areas. Candidate blobs are identified as local maxima in the final response map. In the second stage, intensity and response features are extracted from the candidate blobs and processed through a clustering algorithm to remove false positives, resulting in enhanced accuracy and reliability. The proposed method is evaluated on a medical image dataset and compared against several state-of-the-art methods under various affine transformations. Experimental results demonstrate that the proposed approach outperforms existing methods in terms of blob localization accuracy and affineinvariance.