Learning to see : Image Histogram



Learning to see 

 

An image is what a camera captures. Technically, an image is a discretized reflectance map of an illuminated scene. Light reflecting off objects is captured by a sensor and converted to a discrete value by an analog-to-digital converter (ADC). 

 

Inside the computer memory the image is stored as a 2-dimensional array of numbers. To see, the computer needs to find the information of interest from the 2-dimensional collection of numbers. For example, for autonomous driving application the information of interest would be the obstacle and free lane information in front of the robot. Another example would be optical character recognition (OCR), where the information of interest is the character present in the image.

Histogram

A histogram is a frequency tally i.e. given a collection of objects the histogram tells how many of a each object type are there in the collection. An example follows: 


 An image histogram is similarly a tally of image pixels. For grayscale images, each pixel in the image is a number between 0-255 and the histogram essentially represents the distribution of intensities in the image. 

 The image histogram is a concise yet powerful representation of the image intensity contents. The hills and valleys convey important information about the contents of the image. For example, the hill denoted as 'darker' in the above histogram corresponds to the box shown in the image. For simple images containing a principle foreground against a background this information is enough to extract the foreground object. This is how the green screen background is removed in modern VFX.

Next time we will take a look at an example of object extraction using histogram information. Cheers!

 

 

 

 

 

 

Comments

Post a Comment

Popular posts from this blog

Simulation setup on Win11 and Issac-sim

Image
For setting up isaac-sim (not issac) on my win11 machine I followed the following steps:  (NOTE: You need a recent nvidia GPU for running this simulator, a very powerful card is needed to run this smooth) Download the isaac-sim package from official page at  Download Isaac Sim . I chose the version  standalone@4.5.0-rc.36+release.19112.f59b3005.gl.windows-x86_64 Extract the downloaded package to a directory of your choice. For me it was "E:\sim\isaac_sim" Now there are multiple batch files within in package, which one to choose? The explaination for these scripts is the following  Mostly the first two are relevant for us, but one can try running any of these, it won't break anything.  The main app is the second file "isaac-sim.bat" (note the double 'a'). Based on my prior experience with running `isaac-sim` on Ubuntu, this application works much happily with windows than "Ubuntu". Here's what the app looks like (after taking few minutes to ...
Read more

ISAAC SIM: Stereo camera on a differential drive

Image
 After spending some time online, I found a differential drive robot that comes with a preinstalled stereo camera and a 2D lidar. I do not wish to use the 2D lidar, but I will be using the stereo camera.  Personal rant: Working with Isaac-sim is a test of patience. Have to wait after every other mouse click. Many times, I gave up on the idea of using Isaac-sim because of this reason.  Alright, lets load up an environment (I used the warehouse) and add a jackal robot ( Left Bottom : Isaac Sim Assets --> Robots --> Clearpath --> Jackal ).  Environment : Warehouse and Jackal Robot in Assets  To control this robot, we will need a differential controller ( Tools --> Robotics --> OmniGraph Controllers --> Differential Controller ). Select the jackal robot in "Robot Prim" selection and insert values for wheel radius and wheel distance as (0.098, 4.2). These values are picked as per s uggestion by a user in the nvidia forum . Don't forget to click on ...
Read more

LU decomposition and its special case

Image
Matrices are fundamental objects in modern mathematics and invariably appear in all branches of science. An introductory use case for matrices is in solving linear systems like Ax = b . While this system is introductory and may feel toy'ish, such linear systems are at the center of most engineering problems.  Formulating a system into a linear matrix expression ( Ax = b ) is beyond the scope of this blog (may be covered in future). In this post, we will talk about a matrix factorization technique called $LU$ decomposition and its special case the Cholesky decomposition . These decomposition techniques help in transforming the original matrix expression into a simpler expression that gives solution x of the linear system Ax = b in an easy way. Consider the following two options for matrix $A$ In the case of $A_2$, owing to the upper triangular structure of the matrix solving for x is straightforward via substitution. The LU decomposition provides us with such an upper triangular...
Read more