Assistive device
Affect - based Image Retrieval
Film restoration

Film restoration

Main Researcher: Kyung-tai Kim

Film restoration is to detect the location and extent of detected regions from a given movie film, and if present, to reconstruct the lost information of each regions. Its goal is to achieve a good quality restoration employing a low computing time with the least operator's interaction.

Film restoration is to Film restoration is to detect the location and extent of defected regions from a given movie film, and if present, to reconstruct the lost information of each region`s


  • Film restoration is to achieve a good quality restoration employing a low computing time with the least operator`s interaction.


  • Film restoration has gained increasing attention by many researchers, to support multimedia service of high quality.

Degrade region

  • Old film is degraded by dust, scratch, flick, and so on. Among these, the most frequently degradation is the scratch and blotch.
  • Scratches are usually generated by mechanical rubbings during a film copy and appear in the direction of the film strip on successive frames over the film.
  • Blotches are caused by the presence of particles on the surface of the films like small hairs, dusts, fingerprints and so on.

Film Restoration 

Overview of our system

Film Restoration
Figure 1: An overview of the proposed automatic film restoration.

Figure1 shows an overview of the proposed method, which consists of detection and reconstruction. To accurately detect the degraded regions, we use the textural and shape properties of degradations: scratches are highly contrasted than their neighbors in spatial domain, and appeared as vertical lines, and blotches are discontinuous in spatial and temporal domain. Based on those properties, the degradations are automatically detected, which are passed to the reconstruction stage. The reconstruction is formulated in MRF-MAP (Markov random field - maximum a posteriori) framework, where it is defined as the minimization problem of the posteriori energy function.Then to minimize the function, DGAs are used.


  • The proposed detection method, which consists of candidate detection and verification.
    • Scratches and blotches have the higher contrasts than their neighbors in spatial domain and temporal domain, respectively, thus the candidate regions are first identified by finding the local extreme of a frame along spatial and temporal discontinuities.
    • The verification is performed through texture classification and shape filtering.
      • The textural properties of scratches and blotches are learned using neural networks (NNs)
      • Their shapes are represented using morphological filters.

"Kyung-tai Kim, Eun Yi Kim: Film line strach detection using texture and shape information, Pattern Recognition Letters 31(3): 250-258 (2010)"


In the proposed method, a film reconstruction is formalized in Bayesian approach, so that the reconstruction problem is considered as the minimization problem of the posteriori energy function, E.

where yc(i,j) and yp(i,j)arethe pixels corresponding to a site (i,j) in the current and previous frames, respectively, and the scaling parameters δ, δc and δp are experimentally determined. The E is composed of three terms: the smoothness, and two likelihoods such as the spatial color variance and the temporal color variance. The first imposes the spatial continuity of the region, so that a restoration result with non-homogeneous regions is penalized based on the difference of the pixel. The second term is a constraintas to whetherthe color of a pixel is close to the observed data, that is, it indicates whether or not a pixel is assigned an appropriate color vector. The third term is related to the achievementof temporal continuity of the colors between two consecutive frames in the sequence. The smaller the value of E, the better the restoration.

"Eun Yi Kim, Kyung-tai Kim, Byunggeun Kim: Genetic algorithm-based reconstruction of old films corrupted by scratches and blotches. Pattern Recognition Letters 34(2): 226-237"

  User account

Login / ID:
Forgot Password  |  Registr ]
  Image Battle System

Image Battle Survey System User Rank
N Username Votes Uploads Favourites
1 viplab 43334 0 0
2 changdaekim 27001 0 0
3 myungillee 10001 0 0
4 youjinshin04 3641 0 0
5 myungillee03 3640 0 0
6 changdaekim02 3640 0 0
7 myungillee02 3640 0 0
8 changdaekim03 3640 0 0
9 changdaekim04 3640 0 0
10 myungillee04 3640 0 0
11 myungillee05 3640 0 0
12 myungillee06 3640 0 0
13 changdaekim05 3640 0 0
14 changdaekim01 3640 0 0
15 myungillee01 3640 0 0

Go to Image Battle System page

  Affective Features

Top 10 users are shown

N Username Frequence
1 sanna 4
2 aurshax 2
3 수경 2
4 su-kyung 1
5 Jason 1
6 Jihye Hwang 1
7 EJ 1
8 sodam 1

There are 8 users attended for the survey

Attend to userstudy