Crop/trim image automation 3:4

[Adrian5FF9]

Cropping a lot of images one by one. Need your help to speed up the process.

Final image should be :

Aspect ratio 3:4.

Keep the background color.(might change...but mostly in the 240-255 values)

Center the subject.

Crop above the nose.

Any help is appreciated !

[JJMack]

        var mesh= String.fromCharCode(0, 0, 0, 16, 0, 0, 0, 1, 0, 0, 0, 0, 0, 8, 102, 97, 99, 101, 77, 101, 115, 104, 0, 0, 0, 3, 0, 0, 0, 21, 102, 97, 99, 101, 68, 101, 115, 99, 114, 105, 112, 116, 111, 114, 86, 101, 114, 115, 105, 111, 110, 108, 111, 110, 103, 0, 0, 0, 2, 0, 0, 0, 15, 102, 97, 99, 101, 77, 101, 115, 104, 86, 101, 114, 115, 105, 111, 110, 108, 111, 110, 103, 0, 0, 0, 2, 0, 0, 0, 12, 102, 97, 99, 101, 73, 110, 102, 111, 76, 105, 115, 116, 86, 108, 76, 115, 0, 0, 0, 1, 79, 98, 106, 99, 0, 0, 0, 1, 0, 0, 0, 0, 0, 8, 102, 97, 99, 101, 73, 110, 102, 111, 0, 0, 0, 3, 0, 0, 0, 10, 102, 97, 99, 101, 67, 101, 110, 116, 101, 114, 79, 98, 106, 99, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 110, 117, 108, 108, 0, 0, 0, 2, 0, 0, 0, 0, 88, 32, 32, 32, 100, 111, 117, 98, 63, 222, 208, 81, 126, 75, 23, 229, 0, 0, 0, 0, 89, 32, 32, 32, 100, 111, 117, 98, 63, 214, 68, 141, 240, 18, 52, 86, 0, 0, 0, 13, 102, 101, 97, 116, 117, 114, 101, 86, 97, 108, 117, 101, 115, 79, 98, 106, 99, 0, 0, 0, 1, 0, 0, 0, 0, 0, 13, 102, 101, 97, 116, 117, 114, 101, 86, 97, 108, 117, 101, 115, 0, 0, 0, 5, 0, 0, 0, 5, 115, 109, 105, 108, 101, 100, 111, 117, 98, 63, 204, 40, 245, 192, 0, 0, 0, 0, 0, 0, 8, 117, 112, 112, 101, 114, 76, 105, 112, 100, 111, 117, 98, 191, 202, 225, 71, 160, 0, 0, 0, 0, 0, 0, 8, 108, 111, 119, 101, 114, 76, 105, 112, 100, 111, 117, 98, 63, 202, 225, 71, 160, 0, 0, 0, 0, 0, 0, 10, 109, 111, 117, 116, 104, 87, 105, 100, 116, 104, 100, 111, 117, 98, 63, 196, 122, 225, 64, 0, 0, 0, 0, 0, 0, 11, 109, 111, 117, 116, 104, 72, 101, 105, 103, 104, 116, 100, 111, 117, 98, 63, 204, 40, 245, 192, 0, 0, 0, 0, 0, 0, 20, 102, 101, 97, 116, 117, 114, 101, 68, 105, 115, 112, 108, 97, 99, 101, 109, 101, 110, 116, 115, 79, 98, 106, 99, 0, 0, 0, 1, 0, 0, 0, 0, 0, 20, 102, 101, 97, 116, 117, 114, 101, 68, 105, 115, 112, 108, 97, 99, 101, 109, 101, 110, 116, 115, 0, 0, 0, 0);

That detected this

		var mouth = String.fromCharCode(0, 0, 0, 16, 0, 0, 0, 1, 0, 0, 0, 0, 0, 8, 102, 97, 99, 101, 77, 101, 115, 104, 0, 0, 0, 3, 0, 0, 0, 21, 102, 97, 99, 101, 68, 101, 115, 99, 114, 105, 112, 116, 111, 114, 86, 101, 114, 115, 105, 111, 110, 108, 111, 110, 103, 0, 0, 0, 2, 0, 0, 0, 15, 102, 97, 99, 101, 77, 101, 115, 104, 86, 101, 114, 115, 105, 111, 110, 108, 111, 110, 103, 0, 0, 0, 2, 0, 0, 0, 12, 102, 97, 99, 101, 73, 110, 102, 111, 76, 105, 115, 116, 86, 108, 76, 115, 0, 0, 0, 1, 79, 98, 106, 99, 0, 0, 0, 1, 0, 0, 0, 0, 0, 8, 102, 97, 99, 101, 73, 110, 102, 111, 0, 0, 0, 3, 0, 0, 0, 10, 102, 97, 99, 101, 67, 101, 110, 116, 101, 114, 79, 98, 106, 99, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 110, 117, 108, 108, 0, 0, 0, 2, 0, 0, 0, 0, 88, 32, 32, 32, 100, 111, 117, 98, 63, 223, 253, 143, 32, 0, 0, 0, 0, 0, 0, 0, 89, 32, 32, 32, 100, 111, 117, 98, 63, 195, 208, 61, 128, 0, 0, 0, 0, 0, 0, 13, 102, 101, 97, 116, 117, 114, 101, 86, 97, 108, 117, 101, 115, 79, 98, 106, 99, 0, 0, 0, 1, 0, 0, 0, 0, 0, 13, 102, 101, 97, 116, 117, 114, 101, 86, 97, 108, 117, 101, 115, 0, 0, 0, 1, 0, 0, 0, 5, 115, 109, 105, 108, 101, 100, 111, 117, 98, 63, 169, 153, 153, 160, 0, 0, 0, 0, 0, 0, 20, 102, 101, 97, 116, 117, 114, 101, 68, 105, 115, 112, 108, 97, 99, 101, 109, 101, 110, 116, 115, 79, 98, 106, 99, 0, 0, 0, 1, 0, 0, 0, 0, 0, 20, 102, 101, 97, 116, 117, 114, 101, 68, 105, 115, 112, 108, 97, 99, 101, 109, 101, 110, 116, 115, 0, 0, 0, 0);

detected this:

I think Images Size and the Scale of object in the image  Have a big impact ot the sliders effects in Liquify

[MXKS]

@JJMack  Both Code are used But Mouth is not detected perfectly !

Try this image !

[JJMack]

The mesh I posted  selected most of the lips. I create mesh setting the smile slider  + some small number, The image you posted is quite small like 500x700px when cropped to the image the selection the mesh made looks like this:

[MXKS]

I saw this pic But Lips can't perfect selection of width and height @JJMack 

@jazz-y Can you make a Liquify For Mouth Detection , I have tired too !

[jazz-y]

I still travel and will not be at the computer soon.

Check the level parameters that are applied to the image after the fade is applied in the difference mode. in this thread I have used options that are more suitable for selecting objects with a large area. follow the script step by step and check how well they show the lips. for highlighting small objects I usually use equalize instead of levels.

[JJMack]

I am surprised that these saved meshes work as well as the do,  Images have different Sizes (Number of Pixels), different Aspect Ratios (Canvas Shape) and object  scale for example Heads ( Portrait,  Half Body, Full bode, Groups)   The mesh require for each will different.  I'm sure Liquify will user its AI to detect heads and then have a meshes  for each face that cams be saved, that may work well similar  Images.

Here I saved a mesh  I created toe detect a nose on a square  2K  x 2K 3/4+ body  shot posted in a thread. So  a 4MP image. Here  I show what it select on the original square  3/4 body shot 4MP image then a 1MP portrait,  4mp,  image,  an 8mp Imags,  and a 50MP  image. I have also seen scriptner save different amount a days for mashes.  When I use Jazz-y converter I have to strip most of the beginning mesh.  Sometime Script listen records no mesh seem to leave it undefined for liquify to figure out.  I do not understand how Liquify works here.

[JJMack]

On you small test image this mesh did well.  It does not work well on large images,

        var test = String.fromCharCode(0, 0, 0, 16, 0, 0, 0, 1, 0, 0, 0, 0, 0, 8, 102, 97, 99, 101, 77, 101, 115, 104, 0, 0, 0, 3, 0, 0, 0, 21, 102, 97, 99, 101, 68, 101, 115, 99, 114, 105, 112, 116, 111, 114, 86, 101, 114, 115, 105, 111, 110, 108, 111, 110, 103, 0, 0, 0, 2, 0, 0, 0, 15, 102, 97, 99, 101, 77, 101, 115, 104, 86, 101, 114, 115, 105, 111, 110, 108, 111, 110, 103, 0, 0, 0, 2, 0, 0, 0, 12, 102, 97, 99, 101, 73, 110, 102, 111, 76, 105, 115, 116, 86, 108, 76, 115, 0, 0, 0, 1, 79, 98, 106, 99, 0, 0, 0, 1, 0, 0, 0, 0, 0, 8, 102, 97, 99, 101, 73, 110, 102, 111, 0, 0, 0, 3, 0, 0, 0, 10, 102, 97, 99, 101, 67, 101, 110, 116, 101, 114, 79, 98, 106, 99, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 110, 117, 108, 108, 0, 0, 0, 2, 0, 0, 0, 0, 88, 32, 32, 32, 100, 111, 117, 98, 63, 227, 246, 149, 248, 235, 241, 216, 0, 0, 0, 0, 89, 32, 32, 32, 100, 111, 117, 98, 63, 224, 149, 72, 155, 229, 78, 55, 0, 0, 0, 13, 102, 101, 97, 116, 117, 114, 101, 86, 97, 108, 117, 101, 115, 79, 98, 106, 99, 0, 0, 0, 1, 0, 0, 0, 0, 0, 13, 102, 101, 97, 116, 117, 114, 101, 86, 97, 108, 117, 101, 115, 0, 0, 0, 3, 0, 0, 0, 8, 117, 112, 112, 101, 114, 76, 105, 112, 100, 111, 117, 98, 63, 158, 184, 81, 224, 0, 0, 0, 0, 0, 0, 8, 108, 111, 119, 101, 114, 76, 105, 112, 100, 111, 117, 98, 191, 132, 122, 225, 64, 0, 0, 0, 0, 0, 0, 10, 109, 111, 117, 116, 104, 87, 105, 100, 116, 104, 100, 111, 117, 98, 63, 169, 153, 153, 160, 0, 0, 0, 0, 0, 0, 20, 102, 101, 97, 116, 117, 114, 101, 68, 105, 115, 112, 108, 97, 99, 101, 109, 101, 110, 116, 115, 79, 98, 106, 99, 0, 0, 0, 1, 0, 0, 0, 0, 0, 20, 102, 101, 97, 116, 117, 114, 101, 68, 105, 115, 112, 108, 97, 99, 101, 109, 101, 110, 116, 115, 0, 0, 0, 0);

[MXKS]

Small Image and Big Image will same pixel of mouth detection, then it's perfect @JJMack 

I have requested @jazz-y  For Mouth Detection Of Perfect PiXel, Wait @JJMack  

[JJMack]

I do not believe it will be perfect with vastly different size images. The image you posted is a webs size image 336K pixels 472 px by 712 px.  Here I scaled it up 10 X  34 MP  4720 Px be 7120 Px   about  a 16" by 20" 300ppi print image and I use the same mesh. neither was perfect.  Too little selected on the web image too much selected on the Print Image.  May be perfect on some in between size image.  I have know idea how Adobe Face detection works and the seem to be more than one detection methods used by Adobe. Select Color range has Face detection Liquify has face detection, Neural Filters have face detection and these seem to work differently from what I see selected. Some faces are enevn missed.

Faces vary and perspective proportions vary etc.

[MXKS]

@JJMack  Can you send me perfect code of Mouth Detection ?

[JJMack]

I do not have that mesh any more. It was only I test one I later pasted over it.  Use  Jazz-y converter and create your own.  It may be perfect for a few images it still depends on how Liquify detect faces in images and faces vary in many aspects, Liquify  also fails to detect some faces in Image.  You should not expect perfect selection.   I tend to try to get small selections I then add some area to in scripting code 

[jazz-y]

var mesh = String.fromCharCode(0, 0, 0, 16, 0, 0, 0, 1, 0, 0, 0, 0, 0, 8, 102, 97, 99, 101, 77, 101, 115, 104, 0, 0, 0, 3, 0, 0, 0, 21, 102, 97, 99, 101, 68, 101, 115, 99, 114, 105, 112, 116, 111, 114, 86, 101, 114, 115, 105, 111, 110, 108, 111, 110, 103, 0, 0, 0, 2, 0, 0, 0, 15, 102, 97, 99, 101, 77, 101, 115, 104, 86, 101, 114, 115, 105, 111, 110, 108, 111, 110, 103, 0, 0, 0, 2, 0, 0, 0, 12, 102, 97, 99, 101, 73, 110, 102, 111, 76, 105, 115, 116, 86, 108, 76, 115, 0, 0, 0, 1, 79, 98, 106, 99, 0, 0, 0, 1, 0, 0, 0, 0, 0, 8, 102, 97, 99, 101, 73, 110, 102, 111, 0, 0, 0, 3, 0, 0, 0, 10, 102, 97, 99, 101, 67, 101, 110, 116, 101, 114, 79, 98, 106, 99, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 110, 117, 108, 108, 0, 0, 0, 2, 0, 0, 0, 0, 88, 32, 32, 32, 100, 111, 117, 98, 63, 222, 74, 3, 96, 0, 0, 0, 0, 0, 0, 0, 89, 32, 32, 32, 100, 111, 117, 98, 63, 212, 76, 98, 42, 170, 170, 171, 0, 0, 0, 13, 102, 101, 97, 116, 117, 114, 101, 86, 97, 108, 117, 101, 115, 79, 98, 106, 99, 0, 0, 0, 1, 0, 0, 0, 0, 0, 13, 102, 101, 97, 116, 117, 114, 101, 86, 97, 108, 117, 101, 115, 0, 0, 0, 1, 0, 0, 0, 5, 115, 109, 105, 108, 101, 100, 111, 117, 98, 191, 217, 153, 153, 160, 0, 0, 0, 0, 0, 0, 20, 102, 101, 97, 116, 117, 114, 101, 68, 105, 115, 112, 108, 97, 99, 101, 109, 101, 110, 116, 115, 79, 98, 106, 99, 0, 0, 0, 1, 0, 0, 0, 0, 0, 20, 102, 101, 97, 116, 117, 114, 101, 68, 105, 115, 112, 108, 97, 99, 101, 109, 101, 110, 116, 115, 0, 0, 0, 0);

Liquify only allows you to roughly estimate the area of the face element. This is due to the fact that even the smallest correction can introduce a large amount of noise into the nearest pixels.

Filtering out this noise and leaving only the necessary pixels is a rather creative task that needs to be solved and tested on a large array of images.

In this case, I slightly changed the levels parameters.

lr.levels([10, 0.2, 15])

However, this does not mean that this method will work for all images. Perhaps a gaussian blur should be added before levels to blend the pixels of minimal intensity with the background. 

[r-bin]

Why make such big mouth changes? Try with MouthWidth = 1. Then, through Calculations, create a selection from the Gray Difference. Apply Smooth 5, and... voila, a border without any Levels. As a last resort, you can use Threshold instead of Levels.

[Kukurykus]

You probably refer to script so in your hint there are missing some steps. I understand you duplicate the layer, make liquify for mouth width equal to 1, then calculations (source one: channel: Gray, blending: difference)? What's then to apply smooth to unexisting selection yet?

[jazz-y]

@r-bin, small distortion values on small images can make the difference indistinguishable from noise. Larger values ​​give more noise, but at the same time higher brightness values in places of maximum difference. Since liqufy is essentially a black box, the result of which we can evaluate only indirectly, we always need to look for a balance of settings to solve a specific task.

Also, on complex layouts with a lot of design elements, for no apparent reason, applying difference after liquify sometimes produces large areas of noise in areas not related to the subject's face. That is, one way or another, it is better to provide for ways to remove it.

I prefer gaussian blur as it is a fairly simple function that produces results almost instantly on most computers.

[Kukurykus]

I use smooth that 'kills' unwanted 'single' pixels selection, but it works slow with big value.

[Geppetto Luis]

Jazzy

it works perfectly

thanks for your great contribution.

[Kukurykus]

Loll, he got linked to the topic where it is 🙂

[benjaminh79573840]

Can this be done using pixel ratios instead of 3:4 ?

[Mattia 13]

@jazz-y 

Thank you for posting your solution, it's very helpful!

I'm playing with the script but I'm not very into PS scripting, so maybe you guys can help me 🙂

Basically what I'm trying to do is pass a folder of images into the script, if the image has no face, leave it as is without errors, if the image has face in it, crop it above the nose - as shown by this solution - leaving the rest of the image as is, no subject centering, no left right or bottom crop.

Thank you in advance!

Matt

[brucer18641801]

I have used lighroom for this and it works great - especially when there are alot of images

[Kukurykus]

I confirm, once I installed it and it detects faces really good, at the time of working, so in backround.

[Mattia 13]

How can I use LR to do this?

[Geppetto Luis]

some have noticed that the selection changes based on the dpi of the photo.

[Stephen_A_Marsh]

@Adrian5FF9  (original topic poster)

 MXKS

Please mark one or multiple answers as correct so that others with the same/similar question know that an answer was found.