A Novel Boundary Preserving Adaptive Filter for Improving Superpixel’s Compactness in Dental X-Ray Images
Maxwell A. Teixeira, Domingos B. S. Santos, Thelmo Pontes de Araujo · 2024
Superpixels techniques are commonly used on image segmentation because the objects’s boundaries tend to be captured as a subset of the superpixels’s boundaries. Superpixel’s compactness is an important parameter for the quality of the final segmentation and it may be affected by image noise, which are very common in medical images. X-ray dental images are known to be particularly noisy. This paper analyzes the effects of smoothing and sharpening filters—frequently used in medical image preprocessing—on the compactness of the superpixels, measured by the isoperimetric quotient. Also, some experiments were conducted to validate two hypotheses about the causes of the filtering effects on the compactness of the superpixels. Based on the results, a boundary preserving adaptive filter (with several variants) is proposed, and its performance is compared to eight known filters: mean, median, Gaussian, gradient, Laplacian, morphological top hat, Kuwahara, and Kuwahara-Tomita filters. The proposed filter outperformed the filters tested here in increasing the overall compactness of the superpixels.