Synthesis of possible Smart Cropping strategies with PTGui 5

Best ways to get the crispiest panorama image from a circular fisheye

On the Crop Tab window

Typical possible cropping sizes and shape

Click to enlarge the image


1) 113 degrees is the HFOV of the Sigma EX 8 mm FishEye associated with the Canon EOS 20D DSLR camera. Other APS-C sensors may produce different HFOV with the same lens. Anyhow all present APS-C models (as of Nov 2005) have a HFOV (portrait) of more than 110 degrees (and less than 120 degrees) as all current sensors have a width ranging from 14.8 mm to 15.7 mm for the (Canon EOS 350D) and the (Nikon D70, D2X, Pentax*ist DS and Konica-Minolta 7D) respectively.

2) 110 degrees is the upper limit that the PTGui 5 developer has selected as the acceptable HFOV. After cropping (if any is done) is applied and when the HFOV of at least one image to be stitched is more than 110 degrees, PTGui shall silently pass the warping to be done by PanoTools. PTGUI 5 warp function is then not used.

The EOS 20D CMOS sensor is 15.0 mm of width. It therefore clips the Sigma 8 mm image circle (that is about 21.5 mm of usefull Diameter) along each of the long side of the rectangle. The resulting natural HFOV (portrait mode) is close to 113,3 degrees.

Synthesis table

Most of the possible strategies for stitching and blending six images of a spatial cube (four sides plus Nadir and Zenith) have been tried out.

The success criteria is that the sharper"sweet spot" part of each and every image has been only and entirely taken in account. In short, this happens when the center square of each image matches quite exactly the corresponding face of the "spatial cube".

When peripheral parts of the original image (i.e. coming from the unwanted cropped out areas) are unexpectedly included in the rendered panorama it is considered BAD. Either the Nadir, the Zenith images (or both) are then spoiled by "soft" chunks from the top or bottom part of the four sides. On a close look, it happens that none of the Nadir/Zenith image is left apparent: the four lateral sides creeps up unexpectedly to the poles and cover most of the Nadir/Zenith image if not all!

On the contrary, in some rare instances (with PTGui as the blender) the lateral four faces get spoiled by the Nadir and the Zenith image that are left almost actually "uncropped", these cases were called "barely acceptable". The Nadir and the Zenith images are then a little too prominent over the sides of the cube.

Smartblend may do an seemingly almost good blending (i.e. where every face of the cube is taken into account but really unevenly). BUT to get this less than optimal result one has to make sure the Nadir and Zenith shot are at the first and second (or vice versa) in the list of Source Images. i have called this cases as "Mediocre". If a lateral face is placed on top of the list, it shall spoil the Nadir or Zenit face for a "BAD" result.

Cropping (lens) type PTGui 5 work flow
113x180 Degrees+ (i.e. "No Crop") Full Frame 109x180 Degrees+ Circular 180 Degrees Full Frame Rectangular 113x113 Degrees Circular 113 Degrees Full Frame Rectangular 109x109 Degrees Circular 109 Degrees Warp by Blend by
Barely acceptable Barely acceptable Barely acceptable GOOD GOOD GOOD GOOD PTGui (Warp by PanoTools) PTGui
Unavailable Unavailable Unavailable Unavailable Unavailable Unavailable Unavailable PanoTools (PTStitcher)
>110 Deg >110 Deg >110 Deg >110 Deg >110 Deg GOOD GOOD PTGui (110 Deg HFOV limited)
BAD BAD BAD GOOD GOOD GOOD GOOD PTGui (Warp by PanoTools) Enblend
>110 Deg >110 Deg >110 Deg >110 Deg >110 Deg GOOD GOOD PTGui (110 Deg HFOV limited)
Mediocre or BAD Mediocre or BAD Mediocre or BAD GOOD GOOD GOOD GOOD PTGui (Warp by PanoTools) Smartblend
Unavailable Unavailable Unavailable Unavailable Unavailable Unavailable Unavailable PanoTools (PTStitcher)
>110 Deg >110 Deg >110 Deg >110 Deg >110 Deg GOOD GOOD PTGui (110 Deg HFOV limited)

Using the original six images without any "significant"cropping of both top and bottom parts shall never really give the optimum sharpness on the final panorama.

From this table it appears that GOOD marks are obtained only after cropping is done. I did not anticipate such a clear result from my numerous experiments...

As cropping is therefore recommended, why not go all along the way that makes warp with PTGui 5 really possible (on par with Panorama Tools), by the same token?

To get this, square (i.e. Full Frame fisheye "lens type") or circular (i.e. Circular fisheye "lens type") cropping to keep 109,5 degrees HFOV are the best choices. There is no visible difference in the final rendered panorama between these two different options.


Some figures

To be just below this 110 Deg limit with the 20D/Sigma 8 mm combo, just set the following figures in the Crop Tab window of PTGui for all the images:

From Top to Down: 619 to 2885 pixels (amongst 3504 vertical)

From Left to Right: 35 to 2301 pixels (amongst 2336 horizontal)

These numbers let an effective 2266 x 2266 pixels (either circular or square) area that shall be used for final optimization (the subsequent PTGui supported HFOV shall appear in the Images Parameters" table) and then stitching and blending. This shall result in a 109,6 degrees HFOV or a little less. By the way, the Vertical Field Of View (VFOV) shall be of a same value, meaning that some 40 Degrees of FOV were cropped out on both up and bottom sides, symmetrically from center.

For another Camera than the Canon 20D, a simple linear extrapolation can be made knowing the width of the Sensor (Recall: the EOS 20D CMOS width being 15 mm for 113,3 degrees FOV and 2336 pixels; 2266 pixels give 109.6 Degrees FOV left after cropping)

It's better to not crop (or possibly doing it just near the circle of the original images) before asking PTGUI 5 to generate control points. A preliminary Optimization and trimming of control points can also be performed before cropping, while this is not absolutely necessary. Only after that, you would crop and optimize and then get below the 110 degrees limit...

Cropping first in an external to PTGui Application

As an alternative to the cropping conveniently done within PTGui 5, it is possible to adequately crop the original image in an independant graphic editing application (e.g. Adobe Photoshop) before using those cropped images as "Source Images". Doing so is similar to using a smaller (square or circular) sensor on the camera.

A moderate such a cropping of top and bottom would give a similar situation to using a FullFrame fisheye lens (such as the Nikkor 10,5 mm).

Then the overlap between images is much reduced and it's a pity to not keep these information that provide way to better quality of "Optimization" and CP pairs fitting.

The PTMac case

External cropping described in the previous paragraph is the only way that I have found really feasible in the Apple Mac OS environment.

Compared with PTGui 5, PTMac currently lacks some of the features that were introduced in PTGui. An important example is the support of so-called "negative CP coordinates". A Control Point (CP) has "negative coordinate" when it is located outside of the crop limits. Originally, PTOptimizer (thus PTMac) would not accept these CP and an error message stops the optimization process abruptly.

An Ultimate Way to get perfect Nadir without panorama head and tripod intrusion

I have developped an evolution to the (above) primary Smart Cropping method in order to get rid of these obstructions. It's a direct derivative that can be used together with the five other normally treated faces (i.e. the four lateral sides and Zenith).

Other lenses

Circular fisheye lenses

Of course, the same basic principle could be used for any other lens. Using only the "sweet spot" of it shall yield better final panorama. As an exemple the Peleng 8mm and the Raynox Fisheye converters would greetly benefit from the Smart Crop method.

Full frame fisheye lenses

Full Frame fisheye lens such as the famous Nikkor 10.5 mm Fisheye would benefit from Smart cropping down to square input images keeping the width of the original photographs uncropped.

Full Frame 35mm sensors

Once again the basic principle of Smart cropping that leads to better, contrastier and sharper panorama images could be applied here. Of course the HFOV shall consequently be reduced and it may imply the necessity to shoot more photographs. It is a matter of trade-off to determine which way to go: less photos (e.g. for "action" reporting) or better overall quality.

Overall Conclusion

Some significant cropping is definitely needed in order to get the crispiest possible panorama.

If not applied, some "soft" parts of images shall inevitably spoil the crisp "sweet spot"of others. This may often not be perceived if no side-to-side comparison is done. This is maybe the reason why the cropping function of Panorama Tools (or with another Application) is rather seldomly used.

I ran personnaly recently into a case that started all this technical invertigation. (click on "WHY WOULD YOU SMART-CROP? / A real Case Study" button from the navigation bar at the top).

The six faces should be cropped to maybe about 113 degrees of vertical field of view.

Going below 110 degrees horizontally (HFOV) opens the choice to PTGui 5 internal warping as a possible alternative to Panorama Tools that would be sole possibility otherwise.

Revised 15 November 2005