I'm in love with these lectures on image processing using python. Thank you professor Sreeni! you are doing god's work by providing these excellent lectures for no cost.
dear Sreeni, really nice presentation. we are going to implement gabor filtering in the context of collagen fibers images to train a ML algo. It's been used in some papers with good results, but in no papers is explained well. I have some questions: 1. why you are resizing your kernel to 400,400: I was understanding from the beginning of the video that you set the size accordingly to the size of the features. why not put ksize directly 400x400 instead of resizing? 2. generally for small features of let's say 10 pixels and other features of around 40 pixels. do you think the enhancement is more due to the kernel size or to the sigma size? (let's say all the other parameters fixed) thanks a lot
@20.41.......When you said "So underwhelming"......I was rolling up with laughter......The build-up you gave for Gabor, upped my expectations too......I just stopped to comment, and I have to continue watching it, as to how you will make it to "Exceed the expectations of the huge build-up" you had given till now.....
The problem with horizontal lines is that the filter is much thinner, therefore it find thinner lines, and the kernel is not applied at every pixel (I assume) therefore it divides the lines in the 2 thinner lines
Hello DigitalSreeni, I have to implement a biometric retina recognition system in real time, is it possible with gabor filters for the feature extraction phase? Thanks
Hi its a great intro. thx for that. A small suggestion: it would be a lot better if you had emphasised that x and y in the equation are actually obtained using the angle parameter. Now it is a bit confusing to me. :) cheers
@@DigitalSreeni Sorry for not being clear enough. I meant, when you start explaining what the parameters are of gabor function and iterating over the terms in the equation, you say x square, mentioning that you can also call it x prime without explaining where this x prime comes. Also the theta does not appear in the equation although it is clear that it is a necessary parameter for gabor function. I was suggesting that it would be a lot better if I saw what this x' is explicitly, ofc y' as well.
Thank you for the informative content! What if one wants to detect more than one feature at the same time? Can multiple Gabor kernels be applied simultaneously?
Please watch my next videos in this series to get answers to those questions. In summary, Gabor filters are just convolutional kernels. You can generate lots of these kernels and apply to your images. These filtered images can then be supplied to a classifier that learns about various object representations in the feature images.
I have a small question about using Gabor filters for feature extraction in time series images. So for each image, I followed your tutorial and was able to extract features. However, I want to know how to summarize the values of each feature into a single value (maybe like the mean/median of the values of the Gabor features extracted?), so that I will have a set of single-valued features for each image. I'm new to computer vision so I would appreciate your help a lot....
You can take the average of all pixels in each filtered image and use that as a metric to summarize the filtered image. Also look into GLCM, I did a video on that topic. th-cam.com/video/5x-CIHRmMNY/w-d-xo.html
hey man ... thank you so much for this informative video. Is it possible to tell me how to extract circular features using gabor filter? I have images of bubbles in a pipe, and I want to extract the bubbles.
I’m not sure if Gabor is the efficient approach to detect round objects, you may want to try Hough Circles. Here is a short tutorial: opencv-python-tutroals.readthedocs.io/en/latest/py_tutorials/py_imgproc/py_houghcircles/py_houghcircles.html
Hello hope you doing well. I am working on one project image classification through vision Transformer. I have question related Gabor Filter. when we apply 3D Gabor filter on image is there any gabor weights like(Gaussion Initialization weights) Generate or Not? If it is weights generate so how we can use this weights? Mean to say I want to use 3D Gabor filter weights to initialize hidden layer not with Gaussion initialize weights. Thanks
Thank you for the informative content! I want to get Gabor filter 3*3 size, and I have to change number of ksize? And I want to get Gabor filter in number then, what should I do? can you give the answer?? I am sorry that my English is too bad
There is no single sigma, theta, etc. combination that gives you a filter for edge detection. It depends on the object itself, if it is oriented horizontally then you need to set theta accordingly. This is why you generate banks of Gabor filters to train a machine learning algorithm. Please watch my next few videos to understand the concept better.
It should be whatever I showed in the video, probably 400x400. I record videos live while demonstrating so you can see all parameters right there. I shared the code so you can experiment with filters yourself.
Here is what ChatGPT says about it: A Gabor filter is a type of filter that combines both spatial and frequency information. It is a linear filter that is used to analyze the frequency content of an image in a localized region. The Gabor filter is constructed by modulating a Gaussian kernel with a complex sinusoidal wave, which results in a filter that has both a Gaussian shape in the spatial domain and a sinusoidal shape in the frequency domain. This makes the Gabor filter a spatial-frequency filter that can be used to analyze both the texture and the orientation of an image. In summary, a Gabor filter is a type of filter that incorporates both spatial and frequency information, making it a spatial-frequency filter.
thank you sir for this educative video and clear explanation. just a question, is it possible to use any of these filters (Gabor/Sobel...) in an unsupervised CNN to extract lineaments (exp: faults and fractures) to get better results and accuracy than doing traditional methods ?
If you want to understand the theory better please have a quick look at Wikipedia page for Gabor. You can find other references there. en.wikipedia.org/wiki/Gabor_filter
I'm in love with these lectures on image processing using python. Thank you professor Sreeni! you are doing god's work by providing these excellent lectures for no cost.
These kernel is so powerful in extracting useful features if it was designed successfully!
this is exactly what I am looking for. thank you for the very nice video.
Glad it was helpful!
You are the best. Full information and easy explanation.
You're the best! Fantastic as per usual with this channel, cheers!
Thank you! Cheers!
Dear Professor Sreeni! thank you very much for your great videos!
dear Sreeni, really nice presentation.
we are going to implement gabor filtering in the context of collagen fibers images to train a ML algo. It's been used in some papers with good results, but in no papers is explained well.
I have some questions:
1. why you are resizing your kernel to 400,400: I was understanding from the beginning of the video that you set the size accordingly to the size of the features.
why not put ksize directly 400x400 instead of resizing?
2. generally for small features of let's say 10 pixels and other features of around 40 pixels. do you think the enhancement is more due to the kernel size or to the sigma size? (let's say all the other parameters fixed)
thanks a lot
Excellent explanation .
@20.41.......When you said "So underwhelming"......I was rolling up with laughter......The build-up you gave for Gabor, upped my expectations too......I just stopped to comment, and I have to continue watching it, as to how you will make it to "Exceed the expectations of the huge build-up" you had given till now.....
:)
The problem with horizontal lines is that the filter is much thinner, therefore it find thinner lines, and the kernel is not applied at every pixel (I assume) therefore it divides the lines in the 2 thinner lines
good video for understanding of Gabor filter operation
Glad you liked it
great explanation
Thanks for this amazing video!
Would you like to post some about Log gabor such as their difference and implementation in Python?Thanks a lot!
Hello DigitalSreeni, I have to implement a biometric retina recognition system in real time, is it possible with gabor filters for the feature extraction phase? Thanks
Hi its a great intro. thx for that. A small suggestion: it would be a lot better if you had emphasised that x and y in the equation are actually obtained using the angle parameter. Now it is a bit confusing to me. :) cheers
Not sure what you mean. I watched the video again to see where I made a mistake but couldn’t find any. Can you please elaborate so others can benefit?
@@DigitalSreeni Sorry for not being clear enough. I meant, when you start explaining what the parameters are of gabor function and iterating over the terms in the equation, you say x square, mentioning that you can also call it x prime without explaining where this x prime comes. Also the theta does not appear in the equation although it is clear that it is a necessary parameter for gabor function. I was suggesting that it would be a lot better if I saw what this x' is explicitly, ofc y' as well.
Great video, thanks for the upload. How would you go about detecting veins in a CT scan?
Please watch videos 67 and 67b. If that doesn't work then you need deep learning. So please watch my videos on the topic of U-Net: 73 to 78.
Thank you for the informative content!
What if one wants to detect more than one feature at the same time? Can multiple Gabor kernels be applied simultaneously?
Please watch my next videos in this series to get answers to those questions.
In summary, Gabor filters are just convolutional kernels. You can generate lots of these kernels and apply to your images. These filtered images can then be supplied to a classifier that learns about various object representations in the feature images.
I have a small question about using Gabor filters for feature extraction in time series images. So for each image, I followed your tutorial and was able to extract features. However, I want to know how to summarize the values of each feature into a single value (maybe like the mean/median of the values of the Gabor features extracted?), so that I will have a set of single-valued features for each image. I'm new to computer vision so I would appreciate your help a lot....
You can take the average of all pixels in each filtered image and use that as a metric to summarize the filtered image. Also look into GLCM, I did a video on that topic. th-cam.com/video/5x-CIHRmMNY/w-d-xo.html
@@DigitalSreeni Thank you so much sir...
hey man ... thank you so much for this informative video. Is it possible to tell me how to extract circular features using gabor filter? I have images of bubbles in a pipe, and I want to extract the bubbles.
I’m not sure if Gabor is the efficient approach to detect round objects, you may want to try Hough Circles. Here is a short tutorial: opencv-python-tutroals.readthedocs.io/en/latest/py_tutorials/py_imgproc/py_houghcircles/py_houghcircles.html
@@DigitalSreeni Man you're awesome! Thank you so much!
Hello hope you doing well. I am working on one project image classification through vision Transformer. I have question related Gabor Filter. when we apply 3D Gabor filter on image is there any gabor weights like(Gaussion Initialization weights) Generate or Not? If it is weights generate so how we can use this weights? Mean to say I want to use 3D Gabor filter weights to initialize hidden layer not with Gaussion initialize weights. Thanks
I 😂😂
thank you for ur helpful tutorial. Can u make tutorial showing how to create gabor filter from scratch without using opencv library? thanks
Thankyou so much
What if we have a set of images how can we apply this transformer to all the images ?
Please watch my videos numbered 61 to 68. Here is the link to video 61: th-cam.com/video/BTbIS1mriuY/w-d-xo.html
Thank you for the informative content!
I want to get Gabor filter 3*3 size, and I have to change number of ksize?
And I want to get Gabor filter in number then, what should I do? can you give the answer??
I am sorry that my English is too bad
And can you recommend sigma, theta, lamda, gamma, phi for edge detection? If you think that is difficult I don't care. thank you.
There is no single sigma, theta, etc. combination that gives you a filter for edge detection. It depends on the object itself, if it is oriented horizontally then you need to set theta accordingly. This is why you generate banks of Gabor filters to train a machine learning algorithm. Please watch my next few videos to understand the concept better.
great explanation, can you please guide how to filter an video at particular wavelength for example at 510 nm
510 nm doesn't make sense unless you convert it into pixel units. Please get the pixel size in real life units and define wavelength.
@@DigitalSreeni but how to do that?
May i ask u, why the gabor its become call by open cv?
What is the size of the filter applied to the image at 19:32 seconds? I see that the size of the filter is modified after the application
It should be whatever I showed in the video, probably 400x400. I record videos live while demonstrating so you can see all parameters right there. I shared the code so you can experiment with filters yourself.
@@DigitalSreeni thanks for your reply
@@爱国爱党爱社会主义 applied kernel is 5*5， 400*400 is just for visualization
How can I access to "synthetic.jpg"? any link?
Nice! I am going to try this to remove CT artifacts!
Good luck!
Is gabor filter spatial or frequency filter?
Here is what ChatGPT says about it:
A Gabor filter is a type of filter that combines both spatial and frequency information. It is a linear filter that is used to analyze the frequency content of an image in a localized region.
The Gabor filter is constructed by modulating a Gaussian kernel with a complex sinusoidal wave, which results in a filter that has both a Gaussian shape in the spatial domain and a sinusoidal shape in the frequency domain. This makes the Gabor filter a spatial-frequency filter that can be used to analyze both the texture and the orientation of an image.
In summary, a Gabor filter is a type of filter that incorporates both spatial and frequency information, making it a spatial-frequency filter.
Thank you
What’s job this filter?
Segmentation?
Or enhance image?
thank you
thank you sir for this educative video and clear explanation.
just a question, is it possible to use any of these filters (Gabor/Sobel...) in an unsupervised CNN to extract lineaments (exp: faults and fractures) to get better results and accuracy than doing traditional methods ?
Thankyou so much,,,, Please if you can explain it for MATLAB as well
but why exp function is used?
If you want to understand the theory better please have a quick look at Wikipedia page for Gabor. You can find other references there. en.wikipedia.org/wiki/Gabor_filter
awesome love it
Thank you! Cheers!
Merci !
Thank you very much.
@@DigitalSreeni welcome! Thank you for your channel!
thank you very much...
Welcome!