Classifications

• • H—ELECTRICITY
  • H04—ELECTRIC COMMUNICATION TECHNIQUE
  • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
  • H04N9/00—Details of colour television systems
  • H04N9/79—Processing of colour television signals in connection with recording
  • H04N9/7908—Suppression of interfering signals at the reproducing side, e.g. noise
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
  • G06T5/00—Image enhancement or restoration
  • G06T5/77—Retouching; Inpainting; Scratch removal
• • H—ELECTRICITY
  • H04—ELECTRIC COMMUNICATION TECHNIQUE
  • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
  • H04N1/00—Scanning, transmission or reproduction of documents or the like, e.g. facsimile transmission; Details thereof
  • H04N1/40—Picture signal circuits
  • H04N1/409—Edge or detail enhancement; Noise or error suppression
• • H—ELECTRICITY
  • H04—ELECTRIC COMMUNICATION TECHNIQUE
  • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
  • H04N1/00—Scanning, transmission or reproduction of documents or the like, e.g. facsimile transmission; Details thereof
  • H04N1/40—Picture signal circuits
  • H04N1/409—Edge or detail enhancement; Noise or error suppression
  • H04N1/4097—Removing errors due external factors, e.g. dust, scratches
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
  • G06T2207/00—Indexing scheme for image analysis or image enhancement
  • G06T2207/10—Image acquisition modality
  • G06T2207/10016—Video; Image sequence

Definitions

• the invention relates to a method and an apparatus for
• restoration is carried out digitally after scanning.
• wet scanning techniques which feed the material through a chemical cleansing process to reduce dirt and potential scratches before scanning, they are not used widely due to a number of technical problems which make them quite costly.
• a lot of footage has already been scanned and archived digitally on tape.
• this object is achieved by a method for restoring an image or a sequence of images, which comprises the steps of:
• an apparatus for restoring an image or a sequence of images comprises:
• a second image processor for removing one or more of the detected artifacts from the image or the sequence of images based on the information stored in the metadata database.
• the solution according to the invention which separates the restoration process into detection of objects, e.g. scratch and dirt objects, and removal using an automatic metadata driven workflow, has a plurality of advantages. Contrary to state-of- the art workflows for artifact removal it is not necessary to individually adjust the parameters for different content. At the same time the restoration results are fully reproducible and there is no redundant processing effort due to iterations. Furthermore, the metadata generated during the object detection process allows to generate a comprehensive quality report based on the information stored in the metadata database. Therefore, the solution according to the invention is well-suited for the restoration of large film archives. During the detection process, a metadata database is build automatically with entries for each artifact.
• a quality control process is performed on the information stored in the metadata database after storing information about the detected artifacts in the metadata database and before removing one or more of the detected artifacts from the image or the sequence of images.
• results can be improved through a graphical review of the metadata for each object after detection and before
• the division of the restoration process into a detection process and a removal process allows to modify, add, and remove entries, i.e. objects, within the metadata database during the quality review.
• the quality review takes place exclusively on the metadata database and is based on individual frames or a sequence of frames.
• an artifact is removed from the metadata database by deleting the information about the
• the information about an artifact includes one or more of an index of the artifact, a type of the artifact, coordinates of a rectangular bounding box, a frame index of the appearance of the artifact, a binary pixel mask of bounding box size, a number of affected pixels, a detection weight, a removal flag, the removal toggle flag, and removal information.
• This information enables a comprehensive quality control process.
• the information is well suited for displaying detected defects in a graphical user interface.
• the information about an artifact further includes at least one of an index of the same artifact in a previous image and an index of the same artifact in a subsequent image.
• Objects such as scratch or static dirt usually spread across multiple frames. Therefore, it is useful to store the index of the same object in the frame before and after.
• a graphical user interface which allows to annotate scratch and dirt objects, to individually remove detected scratch and dirt objects, and to add scratch and dirt objects (graphically) by hand.
• the graphical user interface further allows to correct detected artifacts manually using specialized graphical editors.
• artifacts within an image and a total area of the artifacts set for removal within the image is determined. These values are helpful for statistical purposes, but also for the graphical review of the metadata.
• Fig. 1 illustrates the problem of scratch and dirt
• Fig. 2 depicts a frame with two scratches
• Fig. 3 shows the frame of Fig. 2 after removal of the
• Fig. 4 depicts a sequence of frames with a dirt object and a moving object
• Fig. 5 shows the sequence of frames of Fig. 4 after removal of the dirt object
• Fig. 6 illustrates an automatic single step process for
• Fig. 7 shows an operator-assisted frame based restoration process
• Fig. 8 illustrates an operator-assisted sequence based
• FIG. 9 depicts a split restoration process according to the invention
• Fig. 10 shows a further refined three-step workflow for film restoration
• Fig. 11 schematically illustrates a multistep-detection
• Fig. 12 illustrates exemplary object and non-object
• Fig. 13 depicts an exemplary bounding box of a scratch or dirt object
• Fig. 14 shows an exemplary graphical representation of the metadata base
• Fig. 16 depicts an original image without any restoration
• Fig. 17 shows the image of Fig. 16 with metadata annotation subsequent to a detection process
• Fig. 18 depicts an output image after application of a
• the detection of scratch and dirt objects fully automatically, without parameter adjustment by the user, on digitized film footage is a challenging task due to a number of varying factors like noise, grain, flicker, jitter, motion, and content itself.
• the design of detection algorithms generally aims at reducing the misdetection and undetection rates by separating likelihoods around a detection threshold. This is schematically illustrated in Fig. 1.
• the probability density functions of objects usually overlap with probability density functions of other frame content. This means that the object misdetection and undetection rates will always be greater than zero. The reasons for this will be explained in the following.
• Figs. 2 and 3 illustrate the problem of detection and removal of scratches.
• Fig. 2 a frame 1 with two scratches 2, 2' is shown.
• Fig. 3 depicts the same frame 1 after removal of the scratches .
• sequence of frames 1 in Fig. 4 shows a moving object 5, namely a ball, moving from left to right, inside a neighborhood window 6 at frame n.
• the moving object 5 cannot be found in frame n+1 and n-1 at the same window position. The main problem is thus to discriminate dirt from motion. To this end, motion is taken into account.
• Dirt objects 4 can easily be removed by cloning in cases where the content underneath has not changed significantly from frame to frame. The situation is more difficult in the case of
• FIG. 6 an automatic single step restoration process 10 for generating restored content 8 from original content 7 is depicted.
• unsupervised restoration process 10 might be sufficient in many cases once automatic scratch and dirt restoration algorithms perform equally well on any scene and kind of footage. However, perfect algorithms are not yet available. A main issue is to control the algorithm quality for different kinds of footage.
• FIG. 7 A common solution for this problem is shown in Fig. 7, where an operator 9 is placed in the restoration loop to review 11 the results frame by frame and to individually adjust 12 restoration algorithm parameters.
• This iterative single step, fully interactive workflow potentially allows perfect removal of scratch and dirt depending on the time and skills of the operator.
• Sophisticated software tools are available that provide restoration algorithms and help the operator to compare restored content 8 and original content 7.
• the loop is usually split into two stages, as shown in Fig. 8, which illustrates an operator-assisted sequence based restoration process.
• the restoration process iteratively works on sequences of frames, e.g. scene cuts. Restoration is initiated by the operator 9 with a certain parameter set and intermediate results 14 are stored. In a quality control step 13 the intermediate results 14 are either accepted or rejected and processing is restarted with adjusted parameters or bounded processing regions.
• Metadata sets 15 with general information gathered during the restoration stage may be stored. Defect specific information like locations and size may be additionally stored and presented to the operator to guide restoration.
• the generation of metadata allows the rather qualitative process of screening to be turned into a quantitative measurable quality process. For example, the number of objects and the cumulative area (pixel count) per frame is stored as metadata 15 in addition to the restored output 8. After processing, a graphical representation of the complete metadata set gives information about how many pixels per frame have be altered. Still, the generated metadata 15 only help on deciding to accept or reject results.
• the restoration process in order to check the quality of the restoration process before producing the final output, is split into a detection process 20 and a removal process 21, as schematically illustrated in Fig. 9.
• Metadata 15 serves for information transfer between the two processes 20, 21.
• the detection process 20 the detection algorithms work automatically and generate metadata information 15, which is then used during the removal process 21.
• the detection results are aggregated 22 for a comprehensive quality report 23, which may be checked by a human or a machine.
• Fig. 10 depicts a further refined three-step workflow, in which a quality control 24 carried out by the operator 9 is
• the metadata database 15 is edited interactively, e.g. by modifying or adding
• the automatic detection 20 of scratch and dirt objects without manual parameter adjustment, especially on digitized film footage is a complex task due to a number of varying factors like noise, grain, flicker, jitter and (pathological) motion.
• a multistep-detection strategy is applied to scratch and dirt. This strategy is schematically illustrated in Fig. 11.
• a first step 30 the overall noise level is estimated.
• a detection 31 of spurious pixels is performed. By clustering of the spurious pixels potential objects are identified 32.
• weighting function is applied 33 to the potential objects. Subsequently the objects are detected 34, e.g. by weight thresholding. This means that only those
• the metadata database 15 stores potential objects with a weight greater than a defined detection threshold.
• the detection threshold is set relatively low in order to reduce false negative events.
• the objects are validated 34 and flagged 35 for removal.
• the removal flag property is automatically set in the flagging step 35 depending on a predefined removal threshold weight, which is set higher than the detection threshold to prevent false positives from being removed.
• the removal flag also depends on the object validation carried out in the validation step 34.
• the design of the algorithms used for the detection 31 of spurious pixels, identification 32 of potential objects, and application 33 of the weighting function affects the shape of the object and non- object probability density functions. The design aims at reducing their overlap region in order to reduce false
• Fig. 12 illustrates exemplary object and non-object probability density functions and their relation to the
• Scratch and dirt objects are considered as objects of certain type which cover a finite area within a single frame. This is in contrast to artifacts effecting the complete frame such as noise, flicker etc.
• the following set of properties is stored as descriptive information in the metadata database 15:
• y (h,v,t) is a 3-dimensional field with vertical, horizontal, and temporal indices.
• y(h,v, t) may either be a scalar for black and white films or a triple of RGB or YUV values for color film.
• Metadata sets may be divided into object types, e.g. metaScratch (t,n) ,
• metaDirt (t , n) .
• metaX(t,n) is used to denote either one or both types. That means that scratch and dirt objects may have the same number n.
• b (t,n) y (yPos (n) : yPos (n) +yWidth (n) -1 ,
• Fig. 13 An exemplary bounding box is illustrated in Fig. 13.
• indexForward ⁇ !,..., N forward
• indexBackward ⁇ !,..., N backward
• metaX (t , n) contains a field metaX (t , n) .replacement, which provides information of how to carry out removal, i.e. in which way object pixels given by metaX (t , n) .pixel have to be altered or substituted if
• metaX (t , n) . remove, metaX (t , n) . replacement is preferably always defined since metaX (t , n) . remove may be altered during the quality control step 24.
• the field has the advantage that no processing for finding a suitable replacement pattern is required during the removal step 21. This means removal is computationally less demanding, although detection does take longer. Also, the amount of metadata increases.
• the quality control step 24 allows to edit a metadata set 15 that already results in a 'good enough' output to obtain a metadata set 15 for an 'excellent' restoration result. Of course, the actual result will depend on the amount of time spent by the operator 9.
• the quality control step 24 may also be useful to control algorithm quality without modifying data sets manually. Quality control is carried out by checking and modifying metadata. This allows to control the quality more objectively through a countable measure and potentially much quicker than interactive side-by- side frame comparison. After metadata has been approved a quality report for the original file is preferably generated together with a change log for the restored version.
• a graphical representation of the metadata database 15 within a software application with graphical user interface is used.
• An exemplary graphical representation is shown in Figs. 14 and 15.
• the graphical user interface provides aggregated metadata visualization. Outliers indicating false detections can easily be discovered in this way.
• the corresponding frames which are usually few in number, require further inspection.
• scratch and dirt objects 50, 51 that have been detected within a frame are annotated using the bounding box coordinates property.
• different styles are preferably used for the bounding box. Falsely detected objects can be quickly gathered in this way.
• each object 50, 51 is annotated by drawing a rectangular box, according to the size and coordinates stored in metaX (t ,n) .box, on top of the content of the current frame t.
• areaDetected (t) is always equal to or greater than metaX.
• areaRemoved (t) the two bar plots are shown within the same panel 42 with metaX. areaRemoved in front of
• metaX (t) areaRemoved in front of metaX (t) .
• areaDetected is represented by a separate bar having a width greater than one pixel. Since the number of frames I is usually much greater than the horizontal resolution of the display only a section of the plot is visible. This section is chosen in a way that the bar corresponding to the current frame t is always in the center of the plot.
• the corresponding box is clicked or touched in order to toggle the object state of metaX (t , n) . removed and the color and/or dashing of the annotation box accordingly.
• the accumulated area in metaX (t) . areaRemoved is updated instantly, as well as the plots shown in the second panel 42 and the third panel 43.
• the graphical user interface switches to the corresponding frame.
• the content and annotations shown in the first panel 41 as well as the metadata bar plot interval in the third panel 43 are updated instantly.
• a fourth panel 44 allows to select whether only scratches, only dirt, or both shall be displayed in the first panel 41.
• a fifth panel 45 allows to preview the output
• the removal process 21 is applied to the current frame t whenever either the checkboxes for scratch and/or dirt in the fourth panel are active and the view is changed to removed (filtered) by clicking the corresponding selection in the fifth panel 45 or pushing a shortcut key.
• the removal process 21 is applied to the current frame t when a dedicated remove/apply button is pressed (not shown) .
• indexForward ⁇ 1 , N f0rward
• indexBackward ⁇ 1 , N backward ⁇ 0 ⁇ according to f) , metaX (t) .
• areaDetected metaX (t) .
• areaRemoved metaX (t) .
• film restoration is performed by applying metadata metaX (n) to each frame t.
• Figs. 16 to 18 show an exemplary result of a restoration workflow.
• Fig. 16 the original image without any restoration is shown.
• Fig. 17 depicts the same image with metadata
• Fig. 18 shows the output image after application of the removal process 21.

Landscapes

• Engineering & Computer Science (AREA)
• Multimedia (AREA)
• Signal Processing (AREA)
• Physics & Mathematics (AREA)
• General Physics & Mathematics (AREA)
• Theoretical Computer Science (AREA)
• Image Processing (AREA)
• Facsimile Image Signal Circuits (AREA)
• Studio Devices (AREA)

Priority Applications (5)

Applications Claiming Priority (4)

Publications (1)

Family

ID=44509200

Family Applications (1)

Country Status (6)

Cited By (2)

Families Citing this family (1)

Citations (3)

Family Cites Families (8)

• 2011
  • 2011-06-08 EP EP11729075.9A patent/EP2641389A1/en not_active Ceased
  • 2011-06-08 CN CN2011800551140A patent/CN103210636A/zh active Pending
  • 2011-06-08 WO PCT/EP2011/059473 patent/WO2012065759A1/en not_active Ceased
  • 2011-06-08 US US13/884,281 patent/US9167219B2/en not_active Expired - Fee Related
  • 2011-06-08 KR KR1020137012596A patent/KR20130141529A/ko not_active Withdrawn
  • 2011-06-08 JP JP2013539174A patent/JP5873099B2/ja not_active Expired - Fee Related

Patent Citations (3)

Non-Patent Citations (1)

Also Published As

Similar Documents

Legal Events