Abstract
In this article, we presented an audio extraction and restoration system for film optical soundtrack. Different from the traditional audio reproduction methods that extract audio waveform by laser scanning and then do some repair processing in audio domain, we performed noise analysis and restoration algorithms to improve the quality of soundtrack pictures before the audio waveform extraction. By inputting the soundtrack pictures, we first analyze the noise components for each frame, then the restoration algorithms will be carried out, finally we stitch the pictures for a consistent soundtrack to gain the corresponding audio sequence. A database consisted of 19125 soundtrack pictures was built in this paper to evaluate the performance of our framework. The experiment results demonstrated that our framework can achieve similar audio quality as traditional audio extraction method in subject evaluation, meantime show the potential to form the new film restoration technology which can avoid the laser damage.