Blind De-Blurring Scale-Aware Quaternion Core Components and Binary Spiking Network based Keratinocytic Skin Cancer Detection on Facial Images

Illumination texture variability of the skin and image quality is among the challenges in keratinocytic skin cancer detection in facial images. Accuracy is affected by occlusions, low-contrast lesions, and varying skin tones. In addition, model performance is hindered by insufficient data annotations and high computational expense, which requires advanced approaches for robust, reliable, and generalizable detection. Applying the PAD-UFES-20 (Publicly Available Dataset - Universidade Federal do Espírito Santo 2020) dataset, the performance of the Blind DE-Blurring Scale-Aware Quaternion Core Components and Binary Spiking Network (BDEB-SAQCC-BSNet) in keratinocytic skin cancer diagnosis is evaluated. The Blind DE-Blurring Based Light Weight Wiener Filter (BDE-LWWF) is first employed to enhance the quality of the images. The Scale-Aware Meet Transformer (SAMT) is employed for segmentation of impacted areas based on variations. The Planet Optimization Algorithm (Pl-OA) analyses impacted areas in medical images to enhance classification accuracy, whereas the Core Components of Quaternions and Binary Spiking Network (CCQ-BPNet) performs feature extraction for keratinocytic skin cancer classification. The test script in Python utilizes the PAD-UFES-20 dataset, and the results indicate that BDEB-SAQCC-BSNet is superior to other methods with 99.9% efficiency and 99.8% sensitivity, which demonstrates that computer systems can replace human diagnosis.