Table 3 Summary of dermoscopic image preprocessing.

Artifact Removal

This process involves eliminating undesired visual elements such as dark borders, color calibration marks, and image frame shadows. Thresholding and morphological operations were applied to mask these regions, ensuring that the model concentrates only on relevant lesion features without interference from background noise.

Contrast Enhancement

By improving the contrast of the images, we enhanced the visibility of the characteristics of the lesion, thereby supporting the system’s capacity to differentiate between benign and malignant cases.

Hair Removal

To address the presence of hair over lesions, we used a masking-based approach. Hair-like linear patterns were detected using edge-based methods combined with morphological filtering. The detected regions were then smoothed using interpolation, allowing for uninterrupted lesion visibility and a more accurate learning process.

Median Filtering

This technique has been used to reduce the existence of distortion in images. Median filtering is beneficial for retaining the sharp boundaries of objects while omitting insignificant, undesirable elements that may cause confusion for the model.

Resizing to

Resized all images to a fixed input size of \(224 \times 224\) pixels to ensure consistency across the dataset.

Pixel Normalization

The pixel value of each picture, which initially varied from 0 to 255, was scaled down to a range of [0, 1] for stable and consistent training behavior.