Amenabar is a Spanish film director. In one of his movies: 'Abre los ojos' ('Open your eyes'), BTW very poorly remade as an american production under the title of 'Vanilla sky', he managed to close the most crowded avenue in Madrid, Gran Vía, for shooting with no people at all in the scene but the main character.
With this program you can try to do the same, get rid of those annoying tourists, people and cars moving around the building or area you are interested in.
To do that you need to take several pictures of the same scene, using a tripod and manual mode. Set the focus, speed and aperture and not modify them in the whole shooting session.
It's important to take shoots as different from the others as possible, as the program will try to identify the background as those parts that don't change from one picture to another. So the shoots should be spaced in time.
But not too much or you will get differences in the shadows (the sun is moving!), although this can be avoided if using a mask file, explained later.
All the pictures must have exactly the same size in pixels.
The way the program works:
1. Put all images in a separate folder.
2. Choose with '...' any of those files to let the program know where they should be read from.
3. From all the files in 2, choose the one with less moving undesired obstacles, make a copy of it, and paint a pure red (255,0,0) mask over all moving elements that are to be eliminated. The program will only process those areas, leaving all pixels out of the masked zones untouched, so as they appear on this reference masked picture. This file can't be in JPEG (use TIFF or BMP instead) format as red borders must be perfectly delimitated. So if you are going to paint your mask in PS make sure to use a pencil 100% strong, not brush.
This user assistance red mask is optional but highly recommended to:
- Get a much faster processing (6 times or more faster) as only marked areas will be processed.
- Preserve moving parts that BELONG to the desired background (shadows that could move with sun movement, water flows, smoke, ondulating flags,...) from being completely destroyed.
- Avoid undesired artifacts on those areas affected by moving obstacles in may of the pictures, but that are clear on this reference image.
This file can be located in the same directory as the others as will be ignored in the clustering process.
4. Choose a threshold (typ. 5%) for the colour clusters (two pixels will be considered of the same colour if the difference between them is less than this threshold). By changing the threshold better or worse results can be achieved for each set of pictures, try. Generally too low values will produce some sparkle undesired pixels to appear, and too high values will propiciate the averaging of undesired information with the background pixels resulting in "ghost" shapes where moving obstacles were located.
5. Chose a colour model. Absolute is recommended; I am still making tests with the Relative algorithm which should improve background recovery when obstacles are darker than the backgorund (e.g. car tyres over well lighted street concrete).
6. Press 'Remove' and the program will do its calculations (for 17 images of 2Mpx my P4 with 2GB took less than 30 secs to process with mask, nearly 3min unmasked). The result will be saved in the same directory of the pictures with the name: "RemovedFiles17_TH5%_Mask_AbsMode.tif" indicating number of original files and used parameters.
Resulting files will be stored in the same folder as original images, but ignored if more processings are to be done over the same folder (as long as they start with the string "Removed") so you can keep them there and repeat the backgorund recovery using different threshold values for instance.
I expect your feedback.
This is a very early version of the software, I have to improve it a lot. But it works!
This is the simple GUI:
An example: from a series of 17 images in a very crowded avenue:
It is now optional but recommended to assist the program to improve results. We have to choose the image with less obstacles and make a copy of it, setting a pure red (255,0,0) mask over those elements that are to be recovered from the background.
We choose the bottom right image with the red T-shirt boy, and we set roughly (no need to be precise) the mask areas to recover (remove people):
And Amenabar achieved this:
View actual sizeIt works fine.
Tyres of the cars, for being all black, are enemy number one. You can see some black pixels on the street concrete on the right side of the Farmacia.
Also "saboteurs" (characters that stay for a long time, like those people waiting for the bus where the background is impossible to recover) are a problem and may require to be left out of the mask not to get them badly blurred as we did.
But in a sunny day, in a representative building only with tourists, I am sure it will work perfect.
You can download it for free from here (follow Software - Amenabar):
http://perso.wanadoo.es/gluijk
