So the image was corrected as much as possible for the background and whites, and the original artwork on the right was mapped into the photograph: Artwork remapped to conform to the photograph In Conclusion Correcting for the white levels produced a nasty color shift in skin tones and artwork. The original image on the left had an improper white balance, and was only in JPG. Rather than attempting to fix the white balance only, the original artwork was mapped into the photograph to represent the correct colors: Original image on left, original artwork on right Fortunately they still had the high quality artwork used to produce the work being photographed. So the resulting photographs had some very nasty color shifts. One interesting use of this was to correct a problem that someone had where they had taken photos of artwork on display, but didn’t set the white balance properly. Each of the RGB channels of an image can be brought into the software and aligned to help reduce errors caused by TCA in the image. In fact, the open source HDRi software LuminanceHDR uses Hugin behind the scenes for doing the initial alignment before creating the HDRi (align_image_stack).īesides being able to align the image stack for HDRi creation, it also allows for an interesting solution for fixing transverse chromatic aberrations. In this case it is just a pure image alignment. The stitching and alignment of images also works for multiple images of the same scene. If I want parameters that will match the same result as the in-camera JPG, both can be opened in Hugin, and correction parameters can be generated to match the RAW to the JPG distortions. I’ve found this same method to be useful when generating RAW+JPG during a shoot, as the RAW file will be uncorrected. Once the optimization has finished it will produce lens parameter values that can be saved for re-use in the future (or modified for use in other software) to automatically correct lens distortions. The basic idea is to shoot photos of an image that has known straight lines in it, identify points along these lines as straight lines in the software, and to run an optimization against them: Straight lines being identified in Hugin for lens calibration There is a tutorial on the Hugin site that walks through the process. If you start with an image (or images) that have known straight lines in them, they can be used to generate parameters to automatically correct images later. The same technique can also be used in reverse to correct for lens distortion. In the previous example the image was corrected to maintain parallel lines in rendering the final output. Speaking of control lines, the software can also be used to generate lens correction parameters as well. The view on the right is possible without modification using a view camera, but can be approximated using a few extra control lines in the panorama software. This produces an “architectural view” (really axonometric, or really really planometric). From the same panorama, the right image was generated, but in this case the vertical lines of the image were identified in the software. On the left is the rectilinear projection after correcting for lens distortions. The image above is stitched from 3 images. This allows all of the parallel lines to remain parallel in an image: View cameras have the option of shifting the lens up vertically to include the top of a building without having to tilt the film plane. Doing so will lead to converging parallel lines. Shooting a building from ground level for instance will often require the photographer to pan the view up in order to include the top of the building. This comes in particularly handy for shooting architectural subjects where you want to maintain parallel lines in the image. One of the really great capabilities of having the lens plane adjustable relative to the film plane is that the image can be shifted while maintaining orthogonality with the subject. This effect is produced in view cameras by modifying the front plane with the lens relative to the film plane on the back plate.
0 Comments
Leave a Reply. |