Segmentation-Assisted Detection of Dirt Impairments in Archived Film Sequences

Abstract

In this correspondence, a novel segmentation-assisted method for film-dirt detection is proposed. We exploit the fact that film dirt manifests in the spatial domain as a cluster of connected pixels whose intensity differs substantially from that of its neighborhood, and we employ a segmentation-based approach to identify this type of structure. A key feature of our approach is the computation of a measure of confidence attached to detected dirt regions, which can be utilized for performance fine tuning. Another important feature of our algorithm is the avoidance of the computational complexity associated with motion estimation. Our experimental framework benefits from the availability of manually derived as well as objective ground-truth data obtained using infrared scanning. Our results demonstrate that the proposed method compares favorably with standard spatial, temporal, and multistage median-filtering approaches and provides efficient and robust detection for a wide variety of test materials.

Detection of dirt impairments from archived film sequences: survey and evaluations

“…In Ren and Vlachos [44], dirt is detected on the basis of segmented frame images, and then a confidence measurement is proposed for validation and removal of false alarms. This confidence-based validation is also utilized in their proposed adaptive spatial-temporal filtering [45], in which either spatial (such as SSMF or LUM) or motion-compensated filtering like SDIp is applied to each separate image block depending on error residuals after their filtering.…”

“…This confidence-based validation is also utilized in their proposed adaptive spatial-temporal filtering [45], in which either spatial (such as SSMF or LUM) or motion-compensated filtering like SDIp is applied to each separate image block depending on error residuals after their filtering. As this confidence measurement provides a direct clue as dirt, it can be used to remove false alarms in many existing methods [44][45][46][47]. The way how this confidence measurement defines is summarised below.…”

“…, is derived by removing static background from a combined measurement of these raw differences [44][45][46][47], in which…”

“…This assumption even holds true for occluded areas [49], hence image segmentation like regiongrowing seems a good solution for this purpose [44,47]. Besides, false alarms can be also filtered in a combined scheme as discussed below, in which confidence weighting is utilized to the results from other detectors.…”

“…Although good performances may be achieved with these manual degradations, especially in those model-based approaches, they almost certainly will not reflect the real situations hence may lead to unreliable results and evaluations. In Ren and Vlachos [44][45][46], binary masks of ground truth maps are manually obtained from real degraded sequences. However, it is difficult to accurately define the boundary of a dirt cluster due to the fact that dirt contains a hard core but soft boundary [5].…”