57 - How to generate features in Python for machine learning?

Features describe the data you're trying to model. For image processing and machine vision, features can be defined and extracted via digital image filters. This tutorial explains the process of generating features in Python for machine learning purposes.
The code from this video is available at: github.com/bnsreenu/python_fo...

[@gl8218]

Using this example image, this gives a starting point for a new approach to assess wound healing areas and the cell population characteristics around the wound area. Thank you.

[@manuel56354]

Your channel is gold!!

[@samarafroz9852]

You're the best sir doing fantastic job

[@deepthik6828]

I am so happy to watch this tutorial ,after searching a lot I got answer from you.Thank u so much

[@DigitalSreeni]

You are welcome 😊

[@felip6180]

my god, I really LOVED what you showed here

[@DigitalSreeni]

Thank you very much!!!

[@anmolgautam9572]

Thank You Sir.

[@ayaalsalihi6001]

How can we feed several features to specific Classifiers??? And how to extract features from a folder of images Combine different features and than feed it to cnn???

[@aleksandr4369]

Hello Dr. B! I have been watching all your videos and wondered if the copy/pasting you are doing @14:55 would be useful to re-code as a class as you showed in some of your earlier videos. If it is, do you think there would be benefit in creating a tutorial on how to just redo this part _with you excellent coding practice_ as classes? I don't know much programming and am basing this comment off of just having binged your videos! Thank you! But also, you are doing an excellent job and I admire your cross-disciplinary background. I wish these techniques existed and were taught to me in undergrad as I would have started doing all of this much sooner!

[@DigitalSreeni]

For serious programming you can consider using classes but for teaching purposes it confuses newbies. Even many people who code on a regular basis are not good at object oriented programming. Therefore, I stay away from classes although I do use functions sometimes.

[@amineleking9898]

Man, your channel id gold. Thank you very much!

[@manikandanj7891]

Hello Sreeni.. I'm so happy after watching this tutorial.. it gives a detailed explanation on the topic.. thanks for making it.. in data frame the number of rows is 1048576.. is this number is a combination of different features obtained from different filtering techniques?

[@tecfy028]

In this case, the number of rows represents gray values. The image has exactly 1048576 pixels, so there are 1048576 gray values. When the img is loaded with cv2.imread(), it returns a 3-dimensional array (3 arrays one inside the other). The innermost array represents an RGB triplet. You can visualize it like a 2D array (matrix) of RGB triplets in each element entry. RGB triplet -> [Red value, Green value, Blue value] 3D array -> [ [ [Red value, Green value, Blue value], ...], ...] Gray value -> just a number 2D array -> [ [ Gray value, ...], ...] 1D array -> [ Gray value, ...] When the img is grayscaled and then reshaped (-1), the 3-dimensional array turns into a 2-dimensional array (RGB triplet becomes gray value) and then turns into a 1-dimensional array with 1048576 elements. Finally, this 1D array is stored as rows in a data frame. Each data frame column is this 1D array with a filter applied to it, that's why the values shown at 15:45 are different.

[@eneserkuvan]

okay but what about other images ? Do i append them to end of this dataframe ?@@tecfy028

[@immohobot9288]

Nice

[@shabinaa6407]

Sir can you make video on Unet++? I am don't know how to introduce deep supervision in Unet++.

[@HALJUAN1]

Thanks so much for your time and your help with these videos. How can I generate a data frame with these features with all images data set, not just with one? thanks again

[@harpreetdogra222]

Hii there do you got the answer for your query as I have the same question

[@kaveenjayamanna1509]

Although this looks good, the only problem I see is that there will be a lot of rows. For example, a single image could render 1,048,576 rows (1 million rows). So, if I had 200 images, that would be 200 million rows. Is there a way we could derive a coefficient for each image so we end up with just 200 rows per 200 images?

[@amineleking9898]

Thank you man

[@DigitalSreeni]

You're welcome!

[@shuvamgupta2236]

Hi sir. I want to extract geometrical features for example volume of micro organisms in 3d image radius etc. Please make a video on it

[@DigitalSreeni]

Please try regionprops from scikit image. kzbin.info/www/bejne/q2TReGiVes-Zg5I

[@dimitheodoro]

can you make a video for pyradiomics???

[@mahhhhh2599]

Hi sir, does gaussian denoise the image, and cannot be used as features extraction ?

[@DigitalSreeni]

Anything can be used for feature extraction. The whole of point of applying various filters to generate features is that each filter response has different information and one of those will help segment images. So, even Gaussian blurred images contain useful information. Have you ever squinted your eyes to read something that is not very clear? Well, Gaussian smoothing is kind of like that.

[@mahhhhh2599]

@@DigitalSreeni thank you sir, for taking your time answering my question. Much appreciate.

[@abdulhannan-um6to]

Can we do that for lbp as well ?

[@DigitalSreeni]

With LBP you can generate many filters by changing its parameters, for example number of points or radius. So yes, you can do the same with LBP.

[@abdulhannan-um6to]

@@DigitalSreeni Hi can you tell me how we can convert LBP feature vector to featuers to use for machine leraning for multiple images.?

[@TheHarpanOnly]

Why do we need to convert to greyscale first?

[@DigitalSreeni]

Some image processing methods are designed only for single channel (greyscale) images. Therefore, it is safe to convert your images to grey in case color information is not relevant. If you want color information to be included, you can apply the method to each channel and combine them into RGB.

[@TheHarpanOnly]

@@DigitalSreeni i am bit understand what you re talking, since I am new in this matter. I know about doing filter on each RGB channel. But what do you mean "combine" it after filtering? How do we combine each three dataframe to become single dataframe? What will the dataframe look like? Thank you in advance

[@DigitalSreeni]

You may find this video useful which addresses you question. kzbin.info/www/bejne/fYjcnYBri56plac

[@TheHarpanOnly]

@@DigitalSreeni thank you. I'll check it.

[@safakhodabakhsh7822]

entropy filter