Feature Matching (Brute-Force) - OpenCV 3.4 with python 3 Tutorial 26

[@awaisahmad5908]

I have no words for you man. You are best teacher. I have learned alot from these two videos on feature detection and feature matiching.

[@Zainjerr]

Amazing stuff bro I highly appreciate it. You made my life so much easy! God bless you brother

[@yaroslavnasonov3185]

Thank you! The best open cv course!

[@2361244]

Hi Sergio, thanks for the great video. I have a follow up question. How can I use a set of detected key points to find the dissimilarity between two images? Suppose one image had the same features but also extra content. Like in this case your face with the book. How do I draw a bounding box around it

[@pysource-com]

HI rounak, You could use the homography to track the object after you get the features. check this video kzbin.info/www/bejne/f2nXeX-Qeal7qJY

[@sanchaykasturey5334]

Great project!! Is there a way by which we can figure out what is its accuracy

[@javlontursunov6527]

Thank you so much! More power to your elbows!

[@nichtszusagen100]

Thank you again, i have practiced all that .It s amazing

[@hoxyoh2584]

God`s blessing reflected on code, awesome

[@nateb5644]

Is there a method better suited for comparing images of form documents, for example: determining if an image is more similar to a resume rather than a cover letter?

[@shiny_apricot]

a neural network might be a better choice. maybe using a CNN with a binary classifier, or using Yolo by evaluating its results for you own purpose.

[@ywang2000]

how do I match just the whole book cover and get the coordinates of the four corners?

[@aliisdeniz]

Hi. I have a question for something about filtering matches. I did feature matching with ORB in 2 images from taken different cameras at the same time. I just want to eliminate wrong matches between these two. Any idea how can i do that?

[@chocovarela9693]

Hi Sergio:) great video:) thanks how can i know the information about the most significative descriptors?

[@pysource-com]

hi Choco Varela. in this video kzbin.info/www/bejne/f2nXeX-Qeal7qJY i dig deeper into this topic and I explain how to detect and object using the descriptors and keypoints.

[@chocovarela9693]

@@pysource-com thanks:) so i have another question, we have a list sorted "matches" how can i know what descriptor from image 1 (or image 2) corresponding to each one element from list "matches"

[@ValSpiridonov]

Do you think there is a way to detect nasolabial fold adjusstment while producing certain emotions- without neural networks?

[@pysource-com]

I'm not sure about this but it seems pretty hard to me to achieve some reliable results without neural networks.

[@lukasznowarkiewicz]

Thanks for great video!

[@harshavardhanan5269]

hey im facing error while running your code.. matching_result = cv2.drawMatches(img1, kp1, img2, kp2, matches[:50], None, flags=2) cv2.error: OpenCV(3.4.3) C:\projects\opencv-python\opencv\modules\imgproc\src\color.cpp:181: error: (-215:Assertion failed) !_src.empty() in function 'cv::cvtColor' can you please help me out, what is the issue?

[@buianhvu4835]

Thanks for sharing, could you make another video that applies this one for real time object tracking, I find your video is easy to understand.

[@pysource-com]

I'm going to do that as well. already in my plans :)

[@pysource-com]

here you go kzbin.info/www/bejne/f2nXeX-Qeal7qJY :)

[@nikolaigladchenko3182]

Thanks a lot, great video!

[@rob26r]

Nice video - how can I use this to align it to the same angle as the original image?

[@pycad]

Thank you man, really helped me

[@robertobonani8571]

gracias !! you just saved me !! orb e la distanza funziona perfettamente , quindi ? ransac non mi serve piu?

[@fabioarobne]

amazing tutorial

[@geetasowmya2819]

Hi, may I know how to implement SSD for brute force matching to match the keypoint descriptors by myself?

[@johnys369]

it is showing this error Traceback (most recent call last): File "Main.py", line 8, in cv2.imshow("img1", img1) cv2.error: OpenCV(4.1.1) /io/opencv/modules/highgui/src/window.cpp:352: error: (-215:Assertion failed) size.width>0 && size.height>0 in function 'imshow'

[@nghethuatsong]

What is the meaning of "flags=2" in line 17? Thank you. matching_result = cv2.drawMatches(img1, kp1, img2, kp2, matches[:50], None, flags=2)

[@DungPham-ai]

Nice tutorial

[@ayushjain5183]

Hi, I have a doubt in feature extraction. Is it important to convert the image into Grayscale for sift and surf feature extraction?

[@JoelPascualMahindo]

How can I avoid false positives? Thanks in advance.

[@sezermezgil9304]

what is the keypoints exactly what does it indicate ? when i print it out it shows like that

[@haiderissam1479]

Hi Sergio, thanks for the great video. but i have tried to apply SIFT instead of orb but it didn't work, would you please help me by giving me the code of SIFT or SURF. many thanks mentor

[@pysource-com]

Hi, here you can find the code how to apply the sift algorythm: kzbin.info/www/bejne/i4TPZnV-e9ZlfpY here you can see the Feature matching in a live video with sift: kzbin.info/www/bejne/f2nXeX-Qeal7qJY

[@tiaoricardo86]

Hi, You can use SIFT but you must change this part of code cv2.BFMatcher(cv2.NORM_HAMMING) to cv2.BFMatcher(cv2.NORM_L2)

[@samardas1365]

is there any tutorial on gist descriptor?

[@tanzeelhassan2934]

that's awesome, thanks bro

[@Rayskydude]

Love this Bro

[@pw4827]

Have you tested, how well it works with pictures without desirable object? I mean, how much this algorithm is prone to false-positive recognition?

[@pysource-com]

In that case it is going to give you many false results. you could however take from the array matches only value above a certain number, but as I haven't tested it, I can't guarantee the result.

[@pw4827]

Ok. Thanks a lot.

[@bimDe2024]

I super like your explanation, subscribe

[@helaluddinmullah2908]

Please make e a video Sparse representation based python coding. How to train overcomplete dictionary and restore signal or image

[@RabeeQiblawi]

Hi does somebody know how to calculate the overall confidence or accuracy ?

[@pysource-com]

Hi, yes I already did a video, take a look at this series (mostly 2nd and 3rd video): kzbin.info/aero/PL6Yc5OUgcoTlQuAdhtnByty15Ea9-cQly

[@Somfic]

Thanks!

[@Ravithezealous]

can we detect that two images are same or not like face detection, emotion detection.

[@ariel9874]

Thanks

[@harshavardhanachyuta2055]

can we do this using CNN ??

[@robertobonani8571]

is just a mathematical/geometrical process, don't need CNN. convolution could be used for feature extraction but if you need precise matching like I do you can use one convolutional layer to filter some features but no need for training data, is more like a straightforward process. I guess CNN is used more for object recognition and not for matching/tracking. But I have to read the Orb paper.

[@harshavardhanachyuta2055]

@@robertobonani8571 thanks for your reply My question is can we able to detect key points and find the descriptors of the key points using CNN. If the orientation of the image changes we don't get the same key points so i am just thinking is it possible to detect key points with the help neural network.

[@robertobonani8571]

​@@harshavardhanachyuta2055 there are a lot of papers that focus on feature extraction using CNN or other neural networks. it depends on your goal. some features or models are sensitive to orientation, others not. My goal for example is just to match points and calculate the movement of the camera. if your goal is object recognition there are a lot of neural networks. the thing is with neural networks you need a lot of training data so it is also useful to know and use other more geometrical/mathematical procedures.

[@tallurinagapoornima2356]

How can we compare two different size images with different resolution.

[@robertobonani8571]

That's a cool question. I would say is not possible like this. read the orb paper. If you look it produces these parallel straight lines. if would be a different size, it would be possible. but with some preprocessing maybe. I guess the distance would be bigger than

[@javlontursunov6527]

It is not OBR , it is ORB

[@许建-u8l]

OBR OBR ==> Sorry is ORB : )