Image Analytics



What are the techniques available to do analytics on Large Sets of Image Data.

Example : More than 10k Images of statics data available.

How do i apply any model on the data to find out the following things.

  1. Techniques to apply analytics

  2. Libraries for applying models

  3. Need to check the deviation in Images ( Almost every image is similar to other image )

  4. Any tool already available to do the same


Hi @ramgopal_rajkumar, what is the business problem you are facing? What is your aim in doing analytics on image data?


Hi Jai,
I had a Collection of Images of Railway Track between Two KM Range.

If a crack happened between 2.1 KM and 2.2 KM range ,
Right now all the cracks are identified manually by observing the images linearly.
Color deviation will be observed manually.
So i want to apply analytics to observe it systematically by identifying the deviation in Color ( Gradients )

i want to automate it by using any tool or some statistical model


Hi @ramgopal_rajkumar, this seems like a pretty hard problem from a machine learning perspective. If possible, could you post some images of this dataset?

  • How do you propose to explain a “crack” to a computer?
  • Is the data available of the same site or different locations. Because this will severely affect the model.
  • How about time of the image taken? Is it same or different?
  • Could you elaborate this?[quote=“ramgopal_rajkumar, post:3, topic:8983”]
    i want to apply analytics to observe it systematically by identifying the deviation in Color ( Gradients )

What my crude solution would be is to

  • Divide the dataset into two parts, positive examples and negative examples, where positive examples contain images of crack in the railway track and negative examples don’t contain this crack.
  • Apply a machine learning algorithm with appropriate pre-processing to this data.

I also suggest using Deep learning, because they prove to be the best on these kind of problems.

Hope it helps!


On a lighter note, the name’s “JAL”, like the water :wink:


Thanks for your quick response.
I will come up with the replies very soon.

BTW What is that ‘JAL’ ? Could you please more elaborate


@ramgopal_rajkumar My name’s not “JAIFAIZY”, its “JALFAIZY”, i.e. in hindi “pani faizy” :blush: (please don’t ask why!)


Hi Ramgopal - Right now there is a kaggle competition going on that requires image classification State Farm Distracted Driver Detection. If you check the forum of this contest, you’ll find that many people have disclosed their approach. You could use that as a starting point for your project. From what I’ve seen of the competition most people seem to be using deep learning. All the best for making the railways safer for all of us !