anmolaithinker / Boundary-Detection

Task :- Boundary Detection on Unseen Data

I use OpenCV for this task . But Why? I can also use something like object detection

models like YOLO,SSD etc. for this task. The reason is :

• Dataset is not provided for this task. So there must be some reason for that. Reasons can be :
  • We have to find dataset for this task from somewhere.
  • We have to make our own dataset.
  • We have to approach to find the solution for this task from some other method that can be possible to be done in the given time-frame.

3rd Reason seems to be best for me to try this task. So I go for OpenCV for this task.

I am using OpenCV Version : 3.4.3

I have borrowed the images of college IdCard from my friend as my phone camera is not working properly right now.

There are 2 files :

• ExplainOneImage.ipynb -> This file contains all the detailed information how i have done the task with detailed description.

• BulkImages.py -> This python file contains the code for performing the task on all images inside the folder.

To run the code :

Make 2 folders inside the current directory ->

1. images -> Put all your images inside this folder.

2. images-result -> Resultant images will be stored inside this folder.

To run the code use python2

packages that must be installed on python2 virtual enviornment :

• numpy
• PIL
• OpenCV Version: 3.4.3

Steps that I have done in this tasks

First Step Converting image into grayscale image

Second Step to Blurr the image . Why? Reasons:

• Simply to reduce the noise

We can also use medianfilter here infact median filter has advantages over gaussian filter:

• it removes noise while keeping edges relatively sharp.

But let it keep simple and I use gaussian blur filter here

Third Step is to use Thresholding :

Adaptive Thresholding is used why??

• It performs good in all the conditions where image has different lighting conditions in different areas.

Fourth Step is to use Erode and Dilation

For more details on this read on docs :- https://docs.opencv.org/2.4/doc/tutorials/imgproc/erosion_dilatation/erosion_dilatation.html

In Simple Terms :

Erode - It makes the object in white smaller.

Dilate - It makes the object in white bigger.

Fifth Step (Final Step) - It Contains series of Steps

1. Find the Contours from the threshold image . Important thing to note here is :

• cv2.RETR_EXTERNAL => If we use this flag, it returns only extreme outer flags. All child contours are left behind.
• cv2.CHAIN_APPROX_SIMPLE => It will only give us the endPoints.It will compresses horizontal, vertical, and diagonal segments and leaves only their end points.

2. Getting the points of contour which is having the maximum contour area.

3. We will draw the rectangle based on the contours we have find out on an image.

ThankYou