Super Resolution



   I ran across a university site in Israel a decade or so ago. Technion was ahead of its time, at least the computer science department. Among subjects, explored there, was super resolution.
 
   Take a low resolution image and apply processing to get a larger image with the same level of detail.,

   My first instinct was that the task was impossible. You would hasve to manufacture information. The detail in a larger image does not exist, in reality.

   Technion scientists had studied  the problem and developed an approach. This was in March of 2003. I recently ran across some work being done in the area of neural networks that expands the field.
   A paper on SRGAN, images/videos and super resolution attracted my attention. SRGAN is:Super Resolution with Genrative Adversarial Networks.

   This approach is very effective, though some failings in color occur. Structure and form are well done and the manufactured detail is developed from trained associate networks. The failings have to do with some aspects of color.

   The HSV (Hue-Saturation-Value) model is useful for this explanation. The saturation drops in areas. The saturation is not uniform with respect to duplication. The overall Value drops some. The experimental model may benefit from inclusion of the HSV color model in the training set.

   Still, the examples of SR imagery are impressive. One of the uses for SR is a substitute for compression, thus a new way to save disk memory space. A catalog of images could be saved in low resolution format along with the trained model, and recalled when needed. This would be limited to general images where accuracy of detail is not important such as advertising and electronic publications.

Comments

Post a Comment

Popular posts from this blog

HDR history

   As a precursor, let me reiterate, these blogs are for intelligent people with a university background and courses in mathematics, engineering, physics and other of the hard sciences.    The first HDR image was the result of the independent work of Prof. Paul Debevec, U.C., and Gregory Ward Larson, JPL.    Prof. Debevec developed an algorithm to composite several images into one. His work was published in a research paper still available online. In the back of the paper is a Matlab script of the algorithm. Matlab is published by Mathworks, Inc, and is the foremost mathematics laboratory,  Matlab is widely used by scientists and professionals in a science related field.    One of the unknown benefits in HDR is the ability two construct an HDR image with only two exposures (images) if the same film is used for several images. Prof. Debevec derived an additional algorithm for determining the film character and using that data to reduce the number of images needed for a HDRI (High Dy
Read more

HDR : myths and misinterpretations

   I've become aware of widespread  disinformation on HDR. Reviewers of video products and software make claims based on an incomplete and faulty understanding of HDR.    HDR does this, HDR does that, HDR will do this and that.    Many such people do not understand the difference between an HDR image and the compressor that makes the image viewable and printable. The HDR is the attempted storage of  an aspect of 'reality'. This 'reality' is too much for computer displays and printers so someone writes compression software to make the HDR acceptable, i.e., viewable on monitors and printable on printers. There are an infinite number of ways to do this. The HDR image you see and like so much may have been compressed by a different method than the same HDR another person may see, resulting in two visually different images.    This boils down to a matter of interpretation. Much of what is being attributed to HDR images is really the effect of different compressors,
Read more

August 14, 2019 update

A lot has happened since the last post. I have acquired a Konica Minolta MAxxum 5D, replete with strap, charger and rechargable batteries. Everything works...exfept the CCD. All the pictures taken are blacked out. There's no visible physical damage to the unit. I'm considering opening up the unit and perhaps looking for a loose connection caused by a drop of the camera. I located and downloaded  a service manual with diagrams and service data. I want to continue some exploration of HDR imaging. To this end, I combined my interest in my cell phone's cameras with my HDR interest and found a few programs that seem to do a good job at HDR or providing the needed lo-res images for assembling an HDR image. I prefer the assembly method. The tonemap compressor from Radiance, programmed by Eric Reinhard, is the best general purpose compressor I've run across in the years of working, albeit intermittently, with HDR. The apps for the phone are from the Apple AppStore an
Read more