PictureSolve.com
1. Example Restoration of Motion Blurred Picture taken with a hand held camera
Welcome to our web site! PictureSolve.com is providing a service specialized
in extracting latent information from your valuable pictures. We use sophisticated
signal processing methods, to precisely measure a picture degradation,
and then invert it to restore hidden information. The programs we use were originally
developed for Forensic and Law Enforcement applications, and they go beyond image sharpening
and filtering tools offered by commonly available commercial imaging/graphics packages.
BLURS (focus or motion)
Photographs may be blurred by being out of focus, or because of motion during the exposure. This leads to apparent loss of detail. Actually some of the information is irretrievably lost, and some is spread out. Our job is to accurately measure the degradation, and then to retrieve the spread-out information to re-create a sharp picture. This requires specialized tools and knowledge we can provide. Fig 1 above shows what can be achieved if the blur spread function is accurately modeled. In this example an iterative Maximum Entropy algorithm was used on data scanned from a 35mm film negative. It represents the very best that can be done with such a severe blur. Fig 2 below shows an example of a color photo taken of a shelf of books with a digital camera, with severe out of focus. Here a one-pass restoration enables the viewer to read some of the book titles (on the left "The Best of Life", "Drawing Masterclass"), in spite of the very visible reconstruction artefacts. (Some such artefacts are always there, because blurring does destroy some information)
2. Focus Blur Restoration, of books on a shelf
Fig 3 below gives some background in how we go about the complicated business of restoring a picture. The 'Power Spectrum' is a way of displaying the Fourier Transform of the blurred bookshelf image, a complete alternative description. In this form the image shows clearly the nature of the focus blur, with the concentric zero pattern. Notice that in this representation the center shows average brightness, with higher and higher detail in each direction as you move out. A blurred picture attenuates detail at high 'spatial frequencies'. Deblurring involves the controlled inversion of this process. We use another analytic construct called a 'Cepstrum' shown to the right in Fig 3, which helps us to exactly measure the extent of the blur betrayed as a faint ring in the center. Precision is essential in getting any improvement. This is just one of the ways we go about the hardest part of deblurring...measuring the shape and extent of the blur exactly. We also use an innovative interactive dynamic process/display loop to achieve this.
Image deblurring is a much studied subject, and many variations in solution have been found. While it is fairly straightforward to implement an inverse filter, great care has to be used in dealing with the idiosyncracies of the Fourier Transform, in order to prevent unpleasant artefacts in the result. But the most difficult aspect of this field, is the accurate determination of the blur function, or Point Spread Function. We have placed a lot of emphasis on this to make deblurring practical for real world images.
3. Power Spectrum (Magnitude of Fourier Transform) and Cepstrum
Fig 4 below shows a case where the camera was severely out of focus, and the restoration is just good enough to make an intelligent guess at the numbers on the license plate.
4. Focus Blur Restoration of License Plate
Here in Fig 5 is an intriguing example of an attempted out of focus restoration, where just enough improvement is discernable that perhaps one could say from the result that the man was smiling, and that he was probably wearing glasses. Certainly, someone who knows him would now recognize him from the deblurred image.
5. Deblurring unfocused surveillance image
The next picture in Fig 6 shows a restoration quite typical of what is encountered in surveillance
photography. Here there is motion blur, mixed with some focus problems, and the restoration is just sufficient
to bring out some definition in the face.
6. Deblurring Motion Blurred Surveillance photograph
The next picture below Fig 7 is a famous one, from the attempted assassination of Ronald Raegan, where the photo was blurred by
relative motion between subject and camera. The restoration is very typical of what can be achieved in real-world situations.
7. Deblurring curved motion blur in news photo
A major limitation in any attempt at restoration of blurs is image compression (methods to reduce the volume of data).
Much of the imagery seen on the internet is highly compressed,
and this looks like additional noise to the deblurring process. Fig 8 below, shows an attempt to restore a jpeg image.
The restoration looks considerably sharper, and it may be helpful in identifying the vehicle type. Notice how the point spread function shows up
as three red discs in the original, which have been reduced in the restoration to red dots. You can actually watch a live demonstration of this last restoration at
http://www.youtube.com/watch?v=lAKWpoP8JeA/
Wherever possible one should work
with the most original data, preferably uncompressed. However, even here, some new information does become accessible to the viewer.
8. Restoration of jpeg compressed image
The next picture below Fig 8a shows a useful application to deblurring on the web. Here we have a recent news photo (courtesy Reuters) taken during a Senate Hearing, with
the background sharp but the figure in the foreground out of focus.
8a. Original Media Photograph of News event
Fig 8b shows how we've cut out the area of interest, and applied a suitable deblurring filter to bring the face into focus. The result is improved
just sufficiently for recognition. (Secretary of State, Rice). Notice that there are many remaining artefacts and distortions. These are due to the noise in the image, and largely
also to the effects of image compression.
8b. Restoration of selected section
Maybe you have a photo from a bank trap camera, or surveillance equipment where you have a blur,
and you can't go back and take the picture again. These methods can help.
STRUCTURED NOISE (Pattern Removal)
A picture might have a periodic interfering background due to many possible causes. However complicated the problem, this can be remedied by carefully modeling the pattern in the Fourier Domain, and inverting it to clean up the picture. Fig 9 below shows a poor quality TV frame, from which objectionable scan lines are removed.
9. Pattern Removal using tailored Fourier filter to remove scan lines
Fig 10 below, shows a common problem where a halftone screening pattern has been removed. This could well make the difference in recognition from an advertising poster.
10. Removal of Halftone Screening pattern
Here again in Fig 11 is the Power Spectrum of the Halftoned image. The Spectrum shows where the energy of the interfering regularly spaced dots is concentrated, and a carefully tailored filter selectively removes the spots.
11. The filter selectively removes spots at correct spatial frequencies
In Forensic applications, one often finds the need for such filtering
to segregate the desired information (e.g. fingerprint) from an obscuring
background. Fig 12 below shows how Fourier Filtering can be
combined with color enhancement to achieve this. The ridge pattern is lifted
very effectively from a confusing background. This is harder to do than the other examples, but our algorithm is
able to separate globally periodic structures from the features of interest.
12. More complicated pattern discrimination to reveal fingerprint
In Forensic Science, shoe prints and tire marks can be very important evidence. Fig 13 below shows a shoe print on a carpet with a confusing pattern. The matched Fourier filter applied to this separates out the information we want from the background.
13. Fourier Filter to reveal shoe print on carpet
Sometimes video images are acquired in a sub-optimal way either by directly photographing a monitor, or otherwise processing the image through a computer system. If not done properly, the image may well become distorted. However complex it is, if the distortion is periodic, it can be very effectively removed, as in Fig 14 below.
14. Removal of video frame defect with Fourier Filter
Fig 15 below is an example of a degraded video frame from a security surveillance camera, in which two-dimensional patterned noise has somehow been introduced. Our algorithm characterizes the structure of this in the Fourier domain, and separates it from the real data, to yield a much improved result. Note that here, for example one could categorically comment on any pattern the sweater the man is wearing, whereas this would be impossible from the original.
15. Removal of Structured Camera Noise (Courtesy Massmostwanted.org)
You might be surprised how much information is hidden in your degraded picture.
You send us your problem image, and we'll return a restored picture for
a fee.
Pictures can be either monochrome or color.
If you have any comments
or questions, write to Alex at:
Picturesolve@gmail.com
The procedure for submission of materials to PictureSolve are described in These Instructions