Images enhancement (prior applying Pscan)

[George]

Hi all,
what are the standart pre-postprocessing operations do you do to enhance your imagery prior mosaicing?
Could that affect (in negative way) e.g. geometry reconstruction?
George

... and what software do you, guys, use for that?

[RalfH]

I don't use any image enhancement before processing in PhotoScan. Since version 0.9.1, mosaicing is  so good that you usually don't see any edges between images anymore in the final mosaic. As the point finding and matching algorithms are a black box, it is not possible to say how image enhancement would affect these algorithms (unless you did some extensive testing). In any case, if you feel you have to pre-process your images I think it would be best to apply identical pre-processing to all images in your project.

[George]

What then about color correction or even radiometric issues?

[RalfH]

I would apply any colour correction after 3D processing, i.e. in the orthophoto, because during mosaicing PhotoScan also modifies colours (to match colour and brightness between images, deal with vignetting etc.). If you are referring to the potential of orthophotos created by PhotoScan to be used for quantitative analysis which would require radiometrically correct data, I am afraid that this does not sound like a good idea at the moment (at least not for orthos created by mosaicing), because there is no information regarding the changes to the colour values during mosaicing, so we can't be sure that the output would still be radiometrically correct even if we used radiometrically correct input images. Averaging may be better suited for this, as it applies less colour/brightness modification. It would be great if someone from Agisoft could comment on this.

[Wishgranter]

Depend on what you do with the data.... For GIS - its not very good to do anything with colors or contrast or etc - as the bible of Photogrammetry say :-)

But when we work not with medium or large cameras, they are professionaly calibrated - so you can use it for analysis of spectrum ( multiband eventuely )

Our customers devices are better and better over the years, but stil we can do few thigs with that. Best if you shoot in RAWs - much better and sharper results as from JPGs.

A lot of us use it for "scanning" of objects, peoples, buildings, and therefore we can do few things with it. - At least for the whitebalance and better use of RGB space..... For better results.....


AS have mentioned im on the way to put things together , with some testing data for comparsion and etc. So stay tuned, in next few weeks wil dig intro that. Want make complete guide about shooting, PREpost generation of data and etc. even the POSTprocess of some data..... 

[George]

I would apply any colour correction after 3D processing, i.e. in the orthophoto, because during mosaicing PhotoScan also modifies colours (to match colour and brightness between images, deal with vignetting etc.). If you are referring to the potential of orthophotos created by PhotoScan to be used for quantitative analysis which would require radiometrically correct data, I am afraid that this does not sound like a good idea at the moment (at least not for orthos created by mosaicing), because there is no information regarding the changes to the colour values during mosaicing, so we can't be sure that the output would still be radiometrically correct even if we used radiometrically correct input images. Averaging may be better suited for this, as it applies less colour/brightness modification. It would be great if someone from Agisoft could comment on this.

That is interesting view! Thanks Ralf! Ok, 3D processing is quite clear issue (though I have found some, say, issues that e.g. exported DEM gives worse values whereas Point Cloud the best. I still can not find an explanation. The difference was in folds).

When it comes to quantitative analysis e.g. in agriculture. I also a bit doubt about radiometry of the final ortho. However, it would be great to hear Alexey's opinion here.

Btw, what is then about processing of multispectral images (e.g. G, R, NIR) or even (B, G, R, NIR)?

[RalfH]

For multi-channel images (e.g., containing a NIR channel) and/or radiometric applications, I would definitely use averaging or even single image texture mapping to create the orthophoto - as I said, there seems to be quite a bit of modification to the colour/brightness values in mosaicing mode (which is great for visual image quality, but makes things difficult when working with radiometry). When using single image orthophoto texture mapping, you can then use other software to stitch/combine the ortho segments. That's the only way to know where your final colour values come from, and that way you can also check if any modification has occurred in the process (by comparing points in the ortho with the respective points in the input images).

Not sure what you meant with respect to DEM vs. point cloud.

[George]

... which software do you use for stitching of single images then?

Btw, I remember that Pscan had that option for single image texture mapping or sort of ... Or am I wrong?

Under DEM vs Point Cloud I meant Pscan functionality for exporting DEMs and Point Clouds (the results i.e. the Z values derived from those turned to be absolutely different and the Point Cloud  delivers the best quality)

[RalfH]

ENVI would be an option, but much too expensive unless you have it anyway for other reasons. There are probably a number of other remote sensing software products which could be used for this. GlobalMapper would be a cheaper alternative, but I'm not sure about multichannel images there.

PhotoScan has several options to generate orthophotos: mosaic, average and max/min intensity. If you want to create an orthophoto from a single image, after model reconstruction disable all images except the one you want to use and then create an orthophoto in average mode. You can repeat that for all images, export texture each time, and then combine those single image orthophotos in external software.

Regarding the DEM vs. point cloud question, I don't know what is happening there; best to ask Alexey to explain this.