Automatic cropping

I have a cbir system and the output results depends on the cropping of the query image boundary and has to be the same as the cropping of the image boundary in the database. so i was wondering is there a way to crop an image automatically. e.g i do edge detection which gives outlines of the image. i want to crop a certain part of the image automatically. image is a numberplate i want automatic cropping of the image to only include the characters of the number plate and not any other extra little characters unless necessary if the letters mix in with the tiny characters.

an example would be good.

i.e.

this is the start of my code

a=imread(numberplate.jpg);
gr=rgb2gray(a);
bw=im2bw(gr);
edge=edge(bw,'canny',[0.15, 0.3])
crp=imcrop(edge)

This is manual i need automatic.

thanks

Answer 1

[Walter Roberson]

Do you have code now to detect the portion of the image which is the license plate?

Answer 2

[Sean de Wolski]

Copy and paste that into a script||function and it'll be automatic.

[Walter Roberson]

Assuming, of course, that numberplate is a structure with a field named jpg which contains a string which is the name of an image file.

[Sean de Wolski]

Very, true. But he said it works; it just isn't automated.

[Jack]

Exactly it is not automated. the user has to do the cropping manually.

i want an automatic cropping without any user intervention of the boundary.

something like.

loop through the edge detected image where it is black value zero for the boundary around the image and use that as the crop boundary.

[Sean de Wolski]

First off, don't overwrite the MATLAB function EDGE when you name the variable EDGE!

Could you just find the bounding box of the edge image using regionprops and extract it/save it etc? The rest is just manipulating the string for imread. You could do that with a variety of ways but you haven't made anything clear to us as to how what you want automated.

As Walter and I said before; assuming that your above code works (which we agreed it probably doesn't since numberplate.jpg should be in single quotes); everything else IS automated if you put it in a script and run it.

[Jack]

firstly:i was rushing so i forgot the single quotes but that is not the problem.

Sorry for not explaining well

here it is.

input is numberplate image. i want to read it, convert it to gray scale. detect edges using canny method. crop it using imcrop. but the thing i need is the boundary around the image to use in the imcrop function so the user doesn't have to manualy crop the image.

example image 1 - http://www.licenseplatemania.com/fotos/engeland/engeland163.jpg

image2 - http://www.licenseplatemania.com/fotos/engeland/engeland163.jpg

these are the input images they need to be gray scaled and cropped to only include the letters and numbers only minimze any other lines interfering as seen on image 2. i want to find the boundary coordinates to crop.

i hope you understand what i want to crop. my code does work but that only tells the user to manually crop the image i want the cropping process to be automatic

i will try regionprops

[Jack]

i tried using region props but i don't understand how i would get the boundary of interest from the numberplate using regionprops. please help in explaining how to get the boundary coordinates around the number plate.

[Sean de Wolski]

doc regionprops

look at the explanation for BoundingBox.

ceil() the first two entries as the minimum x,y and then add the the third and fourth to the first and second respectively to get the maximum.

Extract with:

Icropped = I(ymin:ymax,xmin:xmax,:);

[Jack]

im really confused i don't understand.

with bounding box i get a lot of coordinates.

i do something like a=regionprops(image);

reg=a.boundingbox;

this results in more than one answer.

is there an example or demo on the regionprops thanks

[Sean de Wolski]

That will give the coordinates for the boundingbox of each object. Read the documentation!!

[Jack]

i did read it, the explanation is only a few lines saying boundbox gives coordinates of xmin ymin and xwidth ywidth.

thats it

here is what i have tried

e=edge(a,'canny',[0.15,0.3]); reg=regionprops(e,'BoundingBox');

reg= 43x1 struct

so i know it has found 43 different bounding coordinates.

i then did reg.BoundingBox

xmin=ceil(ans(1,1)); ymin=ceil(ans(1,2)); xwidth=(ans(1,1)+ans(1,3)); ywidth=(ans(1,2)+ans(1,4));

i get a small area

then i tried after icropped=i(xmin,ymin,xwidth,ywidth) error=too many inputs

help would be appreciated thanks

[Walter Roberson]

At your level of understanding, you should never name a structure followed by a field, only a structure followed by an index followed by a field. reg(K).BoundingBox for K from 1 to length(reg)

You might perhaps be wanting to find the bounding box that encloses all of the bounding boxes you found through edge detection; if so then you should be thinking about min() and max().

There _are_ some useful things you can do with naming a structure name and then a field name with no index, but your first order of duty is to get the code working; optimization can come afterwards.

[Jack]

i did a loop to try and crop the image using all the coordinates from boundingbox.

i wrote

for i=1:length(reg) %in this case length of reg is 43

xmin=reg(k).BoundingBox(1);
ymin=reg(k).BoundingBox(2);
xwidth=reg(k).BoundingBox(3);
ywidth=reg(k).BoundingBox(4);

cr=imcrop(e,'canny',[0.15,0.3]);
imshow(cr);
end

for image 1 when k=4 the whole numberplate was captured including the borders and for image 2 it was when k=3 is it possible to get only the characters not the whole numberplate with the little flag if not i guess it is ok how do i stop the loop when i get that numberplate image boundingbox

note: ceil() the first two entries as the minimum x,y and then add the the third and fourth to the first and second respectively to get the maximum. Extract with: Icropped = I(ymin:ymax,xmin:xmax,:);

do i do ceil of BoundingBox(1)and(2) for x and y min get x and y max by adding BoundingBox(1) to (3) and (2) to (4) respectively

then icrop=i(ymin:ymax,xmin:xmax,:);

is this what was meant by that statement

thanks

[Sean de Wolski]

Assuming that BoundingBox is 1x4 (i.e. it was a 2d image) then this looks correct to me. What results did you attain? Did it work?

[Jack]

but i don't want the user to do any work besides loading image the cropping should be automatic. can someone help. regarding my previous post.

thanks

[Jack]

i tried what you said

when i reach up to reg=regionprops('BoundingBox')

you said to ceil the first two entries i ddn't know which of the 52 coordinates do i look at.

im guessing its is the bound that captures the whole numberplate. through the loop i said.

but when i ceil(first 2 entries i.e. xmin, ymin) then add entries 1 and 2 to 3 and 4 respectively after the ceil func.

its ok up this point but when i do icrop=i(ymin:ymax,xmin:max,:) it says index esceeds matrix dimension or something. if i do Capital I(ymin:ymax,xmin:xmax,:) says undefine function or variable I

i need help with this thanks

[LAXMI NARAYAN SONI]

Yes this is the right way.... U also have to be use a loop.

Answer 3

[Wolfgang Schwanghart]

You might find this function useful: http://www.mathworks.com/matlabcentral/fileexchange/25354-cropmat

HTH, Wolfgang

Answer 4

[priya]

I guess this should work... (not sure though). when you do imcrop() you should be able to right click and get the co-ordinates of the image which you have selected for croping.

%ime1 = findimage();
a=imread(numberplate.jpg);
gr=rgb2gray(ime1);
% BW=im2bw(gr);
edit=edge(gr,'canny',[0.15, 0.3])
im1 = imcrop(edit,[100.5 101.5 456 87]); imshow(im1);

Answer 5

[LAXMI NARAYAN SONI]

if u have a yellow color no. plate then u can use color based cropping... ike u can crop only yellow color. as simple as that.

clc
clear all
close all
a = imread('car3.jpg');
%b=imshow(a)
red = a(:,:,1);
green = a(:,:,2);
blue = a(:,:,3);
figure,imshow(red);
figure,imshow(green);
figure,imshow(blue);
%d = impixel(a);
i=size(a)
for c=0:i

          if a(:,:,1)>13 or <255
              if a(:,:,2)>0 or <201
                  if a(:,:,3)>0 or <104
                    d = imcrop(a);
                    figure,imshow(d)
                  end
              end
          end
  end

[DGM]

It's a good thing that code isn't even valid, because even in concept, it would accomplish nothing worth the cost of execution.

Consider a 1000x1000x3 RGB image. For each element in the array, you ostensibly want to:

1. test to see if the entire image is within a prismatic volume within the RGB color space
2. if it is, display the full image in a window and request that the user interactively select a region with the mouse, right click, and then select "crop image" from a dropdown menu
3. display the cropped region and then discard it

So either the first test fails and the code does literally nothing but spend time warming your CPU, or you're going to make the user do the exact same manual task three million times. Like I said, it's a good thing that it's not valid syntax.

Answer 6

[Samreen kayani]

i want to segment each letter individually but bounding box divid 1 into 11 instead of 1.12 my answere is 11.112