face detection online

I added another feature to the web application. This is opencv's face detection application - just upload an image and see the magic! (click here to go to the application)

Image filtering, smoothing, canny edges

So here is the result of the last post - The current functionality of the application is filtering an image with a given filter and two specific filters - blurring and canny edge detector. Scroll to the right to see the resulting image (or just click here to go to the application)

Image Processing Web Application

I started recently getting into building web applications. Here is a beginner's experience with the issues involved:
My idea was to build some front-end for opencv so people from all around the web could play around with the functionality easily. The current state of my experimentation can be viewed here.
It doesn't use Opencv yet(I'm using ImageMAgick), for reasons I shall explain below.
Here's a list of things you need to set up for such an application:

• A domain - typically these would cost at around 10$ for ordinary domains, I got mine from GoDaddy but there are endless places where you can do it.
• What web application framework you want to use. As I see it the decision should be based on which programming language you feel most comfortable with. Languages like Java, Ruby, Python each have their own framework (I;m familiar with Ruby on rails and recently got introduced to Django for Python).
  
• Hosting environment.When deciding where your application is going to run you should consider the following:

1. What web application framework you are using. You should check that the hosting environment has full support for your framework - this includes looking for reviews and hearing what other people say and not just what the hosting provider says
   
2. What are your computational and network load requirements. As a beginner I use shared hosting which is the cheapest (costs at around 100$/year) since I am not really sure about those issues yet. This means your server is shared with other users. The more expensive option is to have a dedicated server or use a cloud in which resources are dynamically allocated as needed by the application.
3. The customer support offered by the hosting provider. Again, look around the web for reviews.
4. Price
   

So as a starting point I had some minimal knowledge of Ruby on Rails(RoR), a domain I got some time ago from GoDaddy and a shared hosting plan also on GoDaddy (I think they are the cheapest,and for a good reason). This was a really bad experience - do NOT use GoDaddy. Their pages take ages to load and they are horrible in terms of support (at least for RoR - which they claim to support). After struggling for some days I gave up and came across BlueHost.
Up to this point I am extremely pleased with BlueHost. They are perfect as a cheap, shared hosting environment. They support Ruby on Rails, they have loads of documentation, and they have an online live customer support. Their administrator panel (the gui you use to manage you hosting stuff) is very user friendly and load without problem.
The only problem I have with them is the Python support which is rather minimal so I could not get opencv installed properly on their server. So for my experiment I used Rmagick which is a Ruby wrapper for ImageMagick and getting that up and running was rather straight forward without much complications.
My next step would be to get familiar with Django, which is the web application framework for Python, and find a hosting service which supports it (recommendations are welcome). Then continue with my original plan of doing something with Opencv.

Generate a thumbnail of a web page

I found this usefull page which describes how to capture a screenshot of a web page from a given URL. It basically uses Firefox's rendering engine in order to generate a view of the page and then saves it to a JPG.

Sharpening images

This piece of code shows how to sharpen a grayscale image (for color images do the same for each channel).
Sharpening is performed by applying a Laplacian operator on the image and adding the output to the original image.

Finally, this is the operator used (Laplacian+image):

-1 -1 -1

-1 9 -1
-1 -1 -1

(see this nice application in which you can perform convolution with the above operator on a given image)

The code (assumes an image named imname is loadable, final output is stored in image named gray):

gray=cvLoadImage(imname,0)

imwidth = gray.rows

imheight = gray.cols

# convert to 32 bit

gray2=cvCreateImage (cvSize (imheight, imwidth), 32, 1)

for r in range(gray.rows):

for c in range(gray.cols):

gray2[r][c]=gray[r][c]

# define the filter:

lapl=cvCreateImage (cvSize (imheight, imwidth), 32, 1)

m=cvCreateImage (cvSize (3,3), 32, 1)

m[0][0]=-1

m[0][1]=-1

m[0][2]=-1

m[2][0]=-1

m[2][1]=-1

m[2][2]=-1

m[1][0]=-1

m[1][2]=-1

m[1][1]=8

cvFilter2D(gray2,lapl,m)

maxv=0

for r in range(gray.rows):

for c in range(gray.cols):

if(lapl[r][c]>maxv):

maxv=lapl[r][c]

for r in range(gray.rows):

for c in range(gray.cols):

v=int(255*lapl[r][c]/maxv)

gray[r][c]=gray[r][c]+v

maxv=0

for r in range(gray.rows):

for c in range(gray.cols):

if(gray[r][c]>maxv):

maxv=gray[r][c]

for r in range(gray.rows):

for c in range(gray.cols):

v=int(255*gray[r][c]/maxv)

gray[r][c]=v

This example is an application for displaying an image with the ability to stretch the window size arbitrarily (thus enabling to zoom in on the image):


from Tkinter import *
import Image, ImageTk
import sys

#
# an image viewer

class UI(Canvas):

def __init__(self, master, im):


x, y, orig_w, orig_h = im.getbbox()
Canvas.__init__(self, master,width=orig_w, height=orig_h)
tmpim = ImageTk.PhotoImage(im)
self.origim = im
self.image = tmpim
self.create_image(orig_w/2, orig_h/2, image=tmpim)

self.bind('',self.resizeImage)

top=self.winfo_toplevel()
top.rowconfigure(0, weight=1)
top.columnconfigure(0, weight=1)
self.rowconfigure(0, weight=1)
self.columnconfigure(0, weight=1)
self.grid(sticky=N+S+E+W)


def resizeImage(self,event):

im = self.origim.resize((event.width, event.height),Image.ANTIALIAS)
tmpim = ImageTk.PhotoImage(im)
self.image = tmpim
self.create_image(event.width/2, event.height/2, image=tmpim)

if not sys.argv[1:]:

print 'need an image name!'
exit()

else:

filename = sys.argv[1]

root = Tk()
root.title(filename)
im = Image.open(filename)

UI(root, im)
root.mainloop()

converting 8bit image to 32 bit image

If you have an 8bit source image and you want to convert it to a 32 bit image you do it using cvConvertScale:

newim = cvCreateImage (cvSize (src.rows, src.cols, 32, 1)
cvConvertScale(src,newim)

How to crop images with opencv in python

Found this helpful tip from a fellow python-opencv user:
find it here

Opencv example

Here's a short example showing how to use openCV with Python.It reads an image from a file, displays the image, the Harris corner detector on that image and the Canny edge image:
(save this in a file named tmp.py and run with: python tmp.py )

import Image
import os
import sys
from opencv.cv import *
from opencv.highgui import *

def analyzeImage(f,name):

im=Image.open(f)
try:

if(im.size[0]==1 or im.size[1]==1):

return

print (name+' : '+str(im.size[0])+','+ str(im.size[1]))
le=1
if(type(im.getpixel((0,0)))==type((1,2))):

le=len(im.getpixel((0,0)))

gray = cvCreateImage (cvSize (im.size[0], im.size[1]), 8, 1)
edge1 = cvCreateImage (cvSize (im.size[0], im.size[1]), 32, 1)
edge2 = cvCreateImage (cvSize (im.size[0], im.size[1]), 8, 1)
edge3 = cvCreateImage (cvSize (im.size[0], im.size[1]), 32, 3)

for h in range(im.size[1]):

for w in range(im.size[0]):

p=im.getpixel((w,h))
if(type(p)==type(1)):

gray[h][w] = im.getpixel((w,h))

else:

gray[h][w] = im.getpixel((w,h))[0]



cvCornerHarris(gray,edge1,5,5,0.1)
cvCanny(gray,edge2,20,100)

cvNamedWindow("win")
cvShowImage("win", gray);
cvNamedWindow("win2")
cvShowImage("win2", edge1);
cvNamedWindow("win3")
cvShowImage("win3", edge2);

cvWaitKey()

f.close()

except Exception,e:

print e
print 'ERROR: problem handling '+ name

f = open(sys.argv[1],'r')
analyzeImage(f,sys.argv[1])

Image processing tools for python

The most basic library is PIL. it provides pretty basic tools for reading,writing and displaying images, some tools for drawing. In terms of image processing it has some functions for cropping, resizing, sharpening images. For more complex operations I found it insufficient.
The library I found to be useful so far is the Python binding for OpenCV. It can be found here. Installing it should not be too difficult, though I could not get it working until I installed the following:
sudo apt-get install python-opencv

I will describe here my short experience in trying to develop image processing applications using python.

I am developing on Ubuntu, which I highly recommend. For anyone familiar with Linux but is afraid of all the trouble in installing it, Ubuntu is an extremely user friendly environment with an almost Plug-and-Play installation process (you can download it here Ubuntu Download).
For developers linux/ubuntu offers many opensource tools which are easily accessible and ready to use.

The application I am working on involves crawling the web, searching for images, creating a DB of these images and then processing theses images in order to create a useful indexing (you might say it is sort of an image retrieval search engine). It is still in it's very early stages and much of the work is experimental. I will describe here some of the hardships I had to overcome in the process, hopefully it might save some other people's time.

So.... Let's dive right into the WEE BUFETS (Warnings,Errors, Exceptions, Bugs, Unanticipated results, Features, and other Excrements of Toxic Software)

Convert string to file handle:
Here's one sucker I had to deal with yesterday. It goes a little something like this:
Let's say you have a string which is a binary buffer holding the contents of a file. You want to
convert it to a file handle but you don't want to go through the trouble of of writing it to a file and then opening it and getting the file handle. You can do this using the following python code:
import StringIO
f = StringIO.StringIO("string containing file data")

That's it! now you can treat f as a regular file handle much the same way you would after a call to the 'open' function - I encountered this when I saved an image file in a mysql DB as a blob object, then reading from the database.