Using VIPs library and Lua to process pictures

[hzm199822]

To get an original picture from the file, what I want to do is to select a position, cut a rectangle, and make a concave or convex semicircle on the edge of the rectangle, and make this area transparent. Is there an appropriate API or solution?

[jcupitt]

Hi @hzm199822, could you post a sample image showing what you want?

[jcupitt]

For example, this:

#!/usr/bin/luajit

local vips = require "vips"

local im = vips.Image.new_from_file(arg[1], {access = "sequential"})

-- make a transparent square ... just RGBA all 0 (black)
local square = vips.Image.black(64, 64, {bands = 4})

-- insert into the main image
im = im:insert(square, 100, 100)

im:write_to_file(arg[2])

makes a square in an image transparent, eg.:

$ ./try320.lua ~/pics/lion.png x.png

Where do you need the semicircle? Do you need a line, or a segment?

[hzm199822]

![Uploading 1111.png…]()

[hzm199822]

What I want is not all white, but translucency. Or that piece becomes a watermark

[hzm199822]

What I want is that this square is just a benchmark. I want to add a semicircle or subtract a semicircle on any two sides of the four sides of the square. Of course, the diameter of the semicircle must be smaller than the side length

[hzm199822]

What I want is that this square is just a benchmark. I want to add a semicircle or subtract a semicircle on any three sides of the four sides of the square. Of course, the diameter of the semicircle must be smaller than the side length. The translucency on the right side of the image can also be a fuzzy watermark. Then the block on the left side, I hope, is the same as the missing block in the original image, which is equivalent to providing a picture. I need to make two images, one is with watermark, the other is a small piece taken out separately I don't know if luavips can do this for images (add or subtract semicircles on the edges of rectangles)

[jcupitt]

The square in the lion only looks white because the page background is white. Try opening the image in a new tab.

You can overlay complex shapes with composite. You give it an RGBA image, one of the PDF blending modes (eg. over) and a position.

Could you upload some test images? It would save me making them. Just drag the images into the text box.

[hzm199822]

huangzhiming zm.huang@ucloud.cn

签名由 网易邮箱大师 定制

On 11/4/2020 15:58，John Cupittnotifications@github.com wrote：

The square in the lion only looks white because the page background is white. Try opening the image in a new tab. You can overlay complex shapes with composite. You give it an RGBA image and one of the PDF blending modes (eg. over) and a position. Could you upload some test images? It would save me making them. Just drag the images into the text box. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.

[jcupitt]

Composite example:

#!/usr/bin/luajit

vips = require 'vips'

local image = vips.Image.new_from_file(arg[1], {access = "sequential"})
local overlay = vips.Image.new_from_file(arg[2], {access = "sequential"})

-- composite the overlay at 50, 20 using 'over' blend mode
image = image:composite(overlay, "over", {x = 50, y = 20})

image:write_to_file(arg[3])

Generates:

[hzm199822]

I mainly want to extract a figure from the original image, then save it, and change the extracted part of the original image into translucent or other

[hzm199822]

It's important to extract this step, but it seems to me that the API can only extract simple rectangles

[jcupitt]

You can apply masks if you want more complex shapes.

Post a set of sample images and I'll have a go.

[hzm199822]

like this?

huangzhiming zm.huang@ucloud.cn

签名由 网易邮箱大师 定制

On 11/4/2020 17:07，John Cupittnotifications@github.com wrote：

You can apply masks if you want more complex shapes. Post a set of sample images and I'll have a go. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.

[jcupitt]

Use the github website and drag images into the text box.

[hzm199822]

![Uploading 图片.png…]()

[jcupitt]

You need to wait for the upload to finish before pressing Comment.

[hzm199822]

[hzm199822]

OK, now it's OK. My goal is very simple, that is to get such a shape from the original image rather than just a simple rectangle. After saving this shape, I will do some operations on the original image, such as translucency, or other things. I can now extract rectangles and darken them myself, but I don't know if there is such an API to support my goal.

[jcupitt]

Ah, like a jigsaw, I wondered.

Yes, it should be easy. Do you have that shape as a PNG?

[hzm199822]

[hzm199822]

What I want to do is JPG, the encoding format of JPG is simpler than PNG, and the operation time should be shorter

[jcupitt]

I mean, do you have a PNG for the jigsaw piece? It would save me drawing it.

[hzm199822]

[hzm199822]

[hzm199822]

like this?

[jcupitt]

I found this sample:

Example code:

#!/usr/bin/luajit

vips = require 'vips'

local image = vips.Image.new_from_file(arg[1], {access = "sequential"})
local mask = vips.Image.new_from_file(arg[2], {access = "sequential"})

-- the position we place the piece
left = 200
top = 100

-- get just the alpha from the mask
alpha = mask:extract_band(3)

-- image pixels
rgb = image:crop(left, top, mask:width(), mask:height())

-- brighten 50%
rgb = rgb * 1.5

-- attach the mask as the alpha channel
overlay = rgb .. alpha

-- composite back
image = image:composite(overlay, "over", {x = left, y = top})

image:write_to_file(arg[3])

I can run like this:

$ ./try320.lua ~/pics/Gugg_coloured.jpg ~/pics/jigsaw.png x.jpg

To make this:

[hzm199822]

no...

[hzm199822]

What I want to do is to get a jigsaw block from the original image, not a rectangle, because I need this puzzle block to do some operations. This jigsaw block comes from the original picture, and the position obtained from the original image is different each time

[jcupitt]

Sorry, I still don't understand what you want to do.

Are you trying to hide or remove the watermark?

[hzm199822]

[hzm199822]

local vips = require"vips.vips"

local background = "/home/hzm/hzm/vips/8.jpg" local back = vips.Image.new_from_file(background) math.randomseed(tonumber(tostring(os.time()):reverse():sub(1,7))) local s = os.clock() local random_left = math.random(512,914) local random_top = math.random(1,626) local random_length = math.random(81,100) local square_black = vips.Image.black(random_length,random_length,{bands = 3})

local square_back = vips.Image.crop(back,random_left,random_top,random_length,random_length) back = back:insert(square_black,random_left,random_top) back:write_to_file("/home/hzm/hzm/vips/background.jpg") square_back:write_to_file("/home/hzm/hzm/vips/square_back.jpg") local e = os.clock() print("used time:"..e-s.."seconds")

[hzm199822]

thsi api crop Help me get a rectangle from the original, but I want to get a puzzle block instead of a rectangle. What should I do? I've thought about getting rectangles and two semicircles, but I haven't found the API for getting semicircles

[jcupitt]

You just get a rectangle and then apply an alpha to cut the exact shape, like in my example above.

[hzm199822]

I don't think I understand this API properly ：extract_band

[jcupitt]

Example:

#!/usr/bin/luajit

vips = require 'vips'

local image = vips.Image.new_from_file(arg[1])
local mask = vips.Image.new_from_file(arg[2]) 

-- the position we take the piece from
from_left = math.random(0, image:width() - mask:width())
from_top = math.random(0, image:height() - mask:height())

-- the position we put the piece back
from_left = math.random(0, image:width() - mask:width())
from_top = math.random(0, image:height() - mask:height())

-- get just the alpha from the mask
alpha = mask:extract_band(3)

-- image pixels
pixels = image:crop(from_left, from_top, mask:width(), mask:height())

-- brighten and darken 50%
brighter = pixels * 1.5
darker = pixels * 0.5

-- attach the mask as the alpha channel
brighter = brighter .. alpha
darker = darker .. alpha

-- composite back
image = image:composite(darker, "over", {x = from_left, y = from_top})
image = image:composite(brighter, "over", {x = to_left, y = to_top})

image:write_to_file(arg[3])

To make:

I expect you'd like a shadow as well. The easiest way is to draw the shadow on the mask, but you could also compute one with a convolution.

[jcupitt]

If you have a three-band image (eg. RGB) you can extract a band like this:

g = rgb:extract_band(1)

Now g is a one-band image (just the green). Bands number from zero.

[hzm199822]

I'd like to know how you changed the rectangle into this puzzle. Instead of drawing this puzzle in advance, you took it directly from the original picture

[hzm199822]

You mean, let me get a rectangle from the original picture and then turn it into a puzzle?

[hzm199822]

The most important thing is to get it directly from the original drawing, not draw it yourself.

[jcupitt]

I just took the alpha channel from that jigsaw PNG.

You can draw yourself, of course, if you have something like an SVG of the shape.

[hzm199822]

Of course, I know that I can draw by myself, but the effect I want is to take it directly from the original image, because the shape and content are different each time. Do I have to draw again every time? Scholar, can you understand what I mean?

[hzm199822]

[hzm199822]

local vips = require"vips.vips"

local background = "/home/hzm/hzm/vips/8.jpg" local back = vips.Image.new_from_file(background) math.randomseed(tonumber(tostring(os.time()):reverse():sub(1,7))) local s = os.clock() local random_left = math.random(512,914) local random_top = math.random(1,626) local random_length = math.random(81,100) local square_black = vips.Image.black(random_length,random_length,{bands = 3})

local square_back = vips.Image.crop(back,random_left,random_top,random_length,random_length) back = back:insert(square_black,random_left,random_top) back:write_to_file("/home/hzm/hzm/vips/background.jpg") square_back:write_to_file("/home/hzm/hzm/vips/square_back.jpg") local e = os.clock() print("used time:"..e-s.."seconds")

[hzm199822]

I don't believe it. You don't understand what I need

[jcupitt]

Instead of

local square_back = vips.Image.crop(back,random_left,random_top,random_length,random_length)

You can write:

local square_back = back:crop(random_left,random_top,random_length,random_length)

No, I don't understand what you mean :-(

Do you want to detect a jigsaw shape in the original image somehow? Do you know where it is, or do you need to search for it? Is it always the same shape, or do you need to detect any kind of jigsaw shape?

If you can give a clear explanation of what you need, perhaps I can help.

[hzm199822]

The original image is a normal 320 * 160 rectangle. I want to cut the next puzzle shape from the original image. I don't know where I need to cut, because I can use random numbers to determine the coordinates I want to cut. I define their shape as a rectangle and two semicircles, which can be concave or convex. I need to cut it off from the original.

[hzm199822]

[hzm199822]

You know, VIPs provides cuts that I can only find clipping rectangles

[hzm199822]

If there is a semicircle, I think I can cut it twice. The first time I cut the rectangle, the second time I cut two semicircles on the edge of the rectangle, and then I can cut the combination again to achieve the desired effect

[hzm199822]

When I combine these parts, he is a puzzle shape. I then turn the puzzle into black or white or transparent, and put it on the original picture. That's all right