Synthesis algorithm based on overcomplete wavelet expansion for line scratch restoration in old films

Digitalized old films are often degraded by some annoying defects, one of the most common artifacts being line scratches. We present an effective synthesis algorithm for line scratch detection and removal in the overcomplete wavelet expansion (OWE) domain, which is called the OWE-based synthesis algorithm (OSA). OWE is adopted here because of its higher precision in localization than other decimated wavelets. In the detection part, an improved OWE-based cross section method is proposed to find primary line scratches. A morphological refinement method follows, and includes more secondary line scratches. For better scratch removing results, we adopt a pixel-set-based linear interpolation under the criteria of linear minimum square error (LMSE) estimation in every scale. Experimental results show that compared with some classic algorithms, the proposed method can provide higher accuracy and fewer false alarms in line scratch detection, as well as enhanced restoration results.