I have a database of images of one person who is using his hands to show various words and phrases in sign language. The background is white and the only thing changing is the shape of the person's hands and their locations. Now in my gui in matlab, I want the user to be able to choose another image from the same person that was taken at another time doing a sign but wearing the same clothes and then the program will have to compare this against the images in the database and show the most similar. Obviously I can't do pixel by pixel comparison as the images were taken by a hand held mobile camera and slight movement has been inevitable so I should try and locate the hands in the images and compare their shapes. I have no idea how to go about this? I have to say I am new to image processing toolbox in matlab.
Your help is much appreciated
I am doing a phD in computer vision, and I can tell you that it is an unsolved problem. (even in your simple framewrok, with white background)
If you are interested, you might read some works about it ar MIT:
http://people.csail.mit.edu/rywang/handtracking/
or at Oxford:
http://www.robots.ox.ac.uk/~vgg/research/sign_language/index.html
http://www.robots.ox.ac.uk/~vgg/research/hands/index.html
I disagree with you. Such a project can achieve results quickly.
This becomes a problem as soon as the project has to deal with "real life".
Using a single camera, and a completely known background; Opencv provides a simple way to extract hand shape in a image (in about 20 lines of code). You will find plenty of source on the web (have a look at calcbackproj).
After that, what you will have to do is to play with shape, and search for characteristic points.
Begin with some simple signs (example : a circle and a V). How would you recognize one from the other?
There are thousands of papers on sign language; just read the older one to simple ideas flowing :)
Related
I want to implement an application, that is able to recognize pictures from camera input. I don't mean classification of objects, but rather detecting the exact single image from given set of images. So if I for example have an album with 500 pictures, then if I point a camera to one of them, then application will be able to tell it's filename. Most of tutorials I find about CoreML is strictly for image classification (recognizing class of object) and not about recognizing exact image name in camera. This needs to work from different angles as well, and all I can have for training the network is this album with many different pictures (single picture for single object). Can this be somehow achieved? I can't use ARKit Image Tracking, because there will be about 500 of these images, and I need to find at least a list of similar ones first with CoreML / Vision.
I am not sure, but I guess perceptual hashing might be able to help you.
It works in a way that it makes some fingerprint from the reference images, and for a given image, it extracts the fingerprints as well, and then you can find the most similar fingerprints.
in this way, even if the new image is not 100% as the image in the dataset, you still can detect it.
It is actually not very hard to implement. but if you would like, i think phash library is a good one to use.
I want to detect shape and then describe it (somehow) to compare it with server data.
So the first question is, is it possible to detect shape like blob with ARKit?
To be more specific, let's describe my usecase generally.
I want to scan image by phone, get the specific shape, send it on server, compare two images on server (server image is the real one, scanned image would be very similar) and then send back some data. I am not asking about server side, the only question about server side is what should I compare - images using OpenCV, some mathematical description of both images and try to find similarity, etc.).
If the question is hard to understand, let's split it on two easy questions:
1) How to scan 2D object by iPhone and save it (trim the specific shape from its background when object is black and background white).
2) Describe scanned object for comparision with almost the same object.
ARKit has no use here.
You will probably need a lot of CoreImage (for fixing perspective distortion and binarization) and OpenCV logic.
Perhaps Vision can help you a little bit with getting ROI from the entire frame, especially if the waveform image is located in some kind of rectangle.
Perhaps you can train a custom ML model that will recognize specific waveforms or waveforms in general to use with Vision.
In any case, it is not a trivial task.
I need to segment an image in ios for a fashion app by keeping only the foreground image and removing all other background part of the image which should resemble like a tool for removing the background of images in various photo editing tools please help me.
General background subtraction is an unsolved problem, so getting perfect results is going to be a big effort. With that said, you can probably get close. Here are a couple of suggested avenues:
I am guessing that your app will place clothes on a human, or something of the sort. Instead of getting a perfect segmentation, run a person detector, remove all of the image except for the detected person, and fit a part-based human model to the remaining image. Then you have the pose of the person, and can do your image processing accordingly.
Allow the user to input some strokes from the foreground and some strokes from the background, and run a graph-cuts-based image segmentation algorithm on the frame.
Begin your process by having the user not be present in your video stream. From this, learn the background distribution (start with a simple histogram of background pixels, there are much more elaborate schemes but you need a starting place). Then, when the user enters the scene, create a binary image containing the connected components that don't fit into the learned background distribution. This will not be perfect, but you will start to see something close to a binary image where the white pixels are your user, and the black pixels are the background. Use morphology operators to join any large connected components that are slightly separated, and threshold your image to remove small noise in the image, from things like specular objects and illumination changes.
Like I said (and is mentioned in the comments), this is not an easy problem, but you can come up with a good approximation if you put some time into it. I suggest the third method I listed. It is achievable, and can be broken down into small parts so you can tell when you're making progress.
Good luck!
I'm a noob to this forum, but wanted to give it a try.
I'm currently learning Objective-C and Cocoa; trying to build my first iPhone app.
One thing I'm working on is allowing the user to cut his/her face from an image they have taken and paste it into another image. (The idea is cut from one image and paste into another image with a spot for a face to go.)
How can this be done? I am thinking I would allow the user to just touch and drag over their face, in the shape of a rectangle, and then allow them to copy.
Thanks for the help.
Ok, nevertheless your bit arrogant style of asking, here are some guidelines about how to start: generic obj-c/iOS development (start from hello world); UIImage class; camera API; image processing algorithms, face detection algorithms. Go on gradually and do not wish to resolve all problems at once. Write first an application that simple loads an arbitrary photo and shows it to the user. Then modify it that you can crop a specified rectangular area from the image and save it into the new file. Then write an app that switches on the camera that you can take an image and save it to the disk. Then unite what you wrote that you save only a cropped area of the captured image.
When you arrive to this point, you will know much more about software development image handling. AFTER THIS you can start looking for image processing algorithms. Start also here with something simple like a trivial blur filter or similar implemented by you. If you know already a bit of image processing, search for face detection algorithms on the net. It is even possible that you will find some ready framework that includes also these features, or at least you will understand the concepts. You can even come back here to stack overflow and ask for suggestions about a good face detection algorithms, however we still prefer if you have chosen already one and have some concrete issue with it.
I am working on car parking system project. For that, I would like to detect the presence of a car.
Can anybody tell me how I can accomplish this using MATLAB?
Also, what is the algorithm for detecting a car?
There's a whole world of methods for object detection in images. You need to learn a little bit about image processing to solve this problem. I suggest you read about template matching or more generally about Object recognition. Specifically for car detection, if you know they will be seen at a certain angle (head on, for example) i'd try Viola-Jones detection which is implemented in OpenCV as haar-based feature cascade detection. Although OpenCV is not a matlab library, you can probably find something in matlab's image processing toolboxes that does a similar job (or interface into OpenCV)
Background subtraction would be a simple place to start.
In a nutshell:
Can capture an image of your empty parking lot. This is your reference image.
Compare the current image of your parking lot with the reference image. The parts that are different will be of interest.
Problems:
You need to keep updating your reference image to stay current with the conditions (e.g. day, night, cloudy, raining). Sometimes this may not be possible, because your reference image needs to have no cars in it for the approach to work.
Moving things in the background (like trees shaking in the wind) will come up as false positives
Have you considered using 3D/stereoscopic imaging in addition to using 'normal' images? If yes you could open up a whole new world of methods and intelligent tricks to remove objects based upon their distance to the camera. Then, any object that is a certain, fixed distance from the camera (e.g. your background) is easily removable and you can just process the new parts of the image (e.g. cars).
If this interests you I can supply you with an algorithm I have developed to detect animals in a livestock pen, which is a similar concept.