Range Extension for Inpainting Using Image Prediction Based on Non-Harmonic Analysis

This paper introduces a novel scratch image restoration method leveraging non-harmonic analysis to predict and reconstruct damaged regions caused by scratches. Traditional techniques, such as the 2D discrete cosine transform and 2D fast Fourier transform, often rely on harmonic components and periodic assumptions, limiting their ability to handle the complex, non-periodic patterns typical of scratch damage. In contrast, our method identifies and uses non-harmonic components, dynamically adapting to local image features. This approach enhances reconstruction accuracy, particularly for scratched images with intricate structures and non-repetitive textures. Experimental results demonstrate that the proposed method significantly outperforms conventional techniques, achieving average improvements of approximately 10.32 dB in peak signal-to-noise ratio and 0.3795 in structural similarity index measure over 2D discrete cosine transform, and 9.44 dB in peak signal-to-noise ratio and 0.2433 in structural similarity index measure over 2D fast Fourier transform. Our method excels in repairing scratch-affected regions, especially in areas with repetitive structures and distinct characteristics. Our approach provides a promising solution for accurate and reliable scratch image restoration.