I am trying to do a number of things via MATLAB but I am getting a bit lost with what techniques to use. My ultimate goal is to extract various measurements from a users fingerprint presentation, e.g. how far the finger over/undershoots, the co-ordinates of where the finger enters, the angle of the finger.
In my current setup, I have a web camera recording footage of a top down view of the presentation which I then take the video file and break down into individual frames. https://www.dropbox.com/s/zhvo1vs2615wr29/004.bmp?dl=0
What I am trying to work on at the moment is using ROI based image processing to create a binary mask around the edges of the scanner. I'm using the imbw function to get a binarised image and getting this as a result. https://www.dropbox.com/s/1re7a3hl90pggyl/mASK.bmp?dl=0
What I could use is some guidance on where to go from here. I want to be able to take measurements from the defined ROI to work out various metrics e.g. how far a certain point is from the ROI so I must have some sort of border for the scanner edges. From my experience in image processing so far, this has been hard to clearly define. I would like to get a clearer image where the finger is outlined and defined and the background (i.e. the scanner light/blocks) are removed.
Any help would be appreciated.
Thanks
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 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 :)
How to implement a way to measure distances in real time (video camera?) on the iPhone, like this app that uses a card to compare the size of the card with the actual distance?
Are there any other ways to measure distances? Or how to go about doing this using the card method? What framework should I use?
Well you do have something for reference, hence the use of the card. Saying that after watching the a video for the app I can't seem it seems too user friendly.
So you either need a reference of an object that has some known size, or you need to deduct the size from the image. One idea I just had that might help you do it is what the iPhone's 4 flash (I'm sure it's very complicated by it might just work for some stuff).
Here's what I think.
When the user wants to measure something, he takes a picture of it, but you're actually taking two separate images, one with flash on, one with flash off. Then you can analyze the lighting differences in the image and the flash reflection to determine the scale of the image. This will only work for close and not too shining objects I guess.
But that's about the only other way I thought about deducting scale from an image without any fixed objects.
I like Ron Srebro's idea and have thought about something similar -- please share if you get it to work!
An alternative approach would be to use the auto-focus feature of the camera. Point-and-shoot camera's often have a laser range finder that they use to auto-focus. iPhone doesn't have this and the f-stop is fixed. However, users can change the focus by tapping the camera screen. The phone can also switch between regular and macro focus.
If the API exposes the current focus settings, maybe there's a way to use this to determine range?
Another solution may be to use two laser pointers.
Basically you would shine two laser pointers at, say, a wall in parallel. Then, the further back you go, the beams will look closer and closer together in the video, but they will still remain the same distance apart. Then you can easily come up with some formula to measure the distance based on how far apart the dots are in the photo.
See this thread for more details: Possible to measure distance with an iPhone and laser pointer?.
I have a need to measure a room (if possible) from within an iPhone application, and I'm looking for some ideas on how I can achieve this. Extreme accuracy is not important, but accuracy down to say 1 foot would be good. Some ideas I've had so far are:
Walk around the room and measure using GPS. Unlikely to be anywhere near accurate enough, particularly for iPod touch users
Emit sounds from the microphone and measure how long they take to return. There are some apps out there that do this already, such as PocketMeter. I suspect this would not be user friendly, and more gimmicky than practical.
Anyone have any other ideas?
You could stand in one corner and throw the phone against the far corner. The phone could begin measurement at a certain point of acceleration and end measurement at deceleration
1) Set iPhone down on the floor starting at one wall with base against the wall.
2) Mark line where iPhone ends at top.
3) Pick iPhone up and move base to where the line is you just drew.
4) Repeat steps 1->3 until you reach the other wall.
5) Multiply number of lines it took to reach other wall by length of iPhone to reach final measurement.
=)
I remember seeing programs for realtors that involved holding a reference object up in a picture. The program would identify the reference object and other flat surfaces in the image and calculate dimensions from that. It was intended for measuring the exterior of houses. It could follow connected walls that it could assume were at right angles.
Instead of shipping with a reference object, as those programs did, you might be able to use a few common household objects like a piece of printer paper. Let the user pick from a list of common objects what flat item they are holding up to the wall.
Detecting the edges of walls, and of the reference object, is some tricky pattern recognition, followed by some tricky math to convert the found edges to planes. Still better than throwing you phone at the far wall though.
Emit sounds from the microphone and measure how long they take to return. There are some apps out there that do this already, such as PocketMeter. I suspect this would not be user friendly, and more gimmicky than practical.
Au contraire, mon frère.
This is the most user friendly, not to mention accurate, way of measuring the dimensions of a room.
PocketMeter measures the distance to one wall with an accuracy of half an inch.
If you use the same formulas to measure distance, but have the person stand near a corner of the room (so that the distances to the walls, floor, and ceiling are all different), you should be able to calculate all three measurements (length, width, and height) with one sonar pulse.
Edited, because of the comment, to add:
In an ideal world, you would get 6 pulses, one from each of the surfaces. However, we don't live in an ideal world. Here are some things you'll have to take into account:
The sound pulse causes the iPhone to vibrate. The iPhone microphone picks up this vibration.
The type of floor (carpet, wood, tile) will affect the time that the sound travels to the floor and back to the device.
The sound reflects of off more than one surface (wall) and returns to the iPhone.
If I had to guess, because I've done something similar in the past, you're going to have to emit a multi-frequency tone, made up of a low frequency, a medium frequency, and a high frequency. You'll have to perform a fast Fourier Transform on the sound wave you receive to pick out the frequencies that you transmitted.
Now, I don't want to discourage you. The calculations can be done. However, it's going to take some work. After all PocketMeter has been at it for a while, and they only measure the distance to one wall.
I think an easier way to do this would be to use the Pythagorean theorem. Most rooms are 8 or 10 feet tall and if the user can guess accurately, you can use the camera to do some analysis and crunch the numbers. (You might have to have some clever way to detect the angle)
How to do it
I expect 5 points off of your bottom line for this ;)
Let me see if it helps. Take an object of known length and keep it beside the wall and with Iphone, take pic of wall along with the object that you kept beside the wall. Now get the ratio of wall width and object width from the image in Iphone. And as you know the width of the object, you can easily calcualte the width of wall. repeat it for each wall and you will have a room measurement.
Your users could measure a known distance by pacing it off, and thereby calibrate the length of their pace. Then they could enter the distance of each wall in paces, and the phone would convert it to feet. This would probably be very convenient, and would probably be accurate to within 10%.
If they may need more accurate readings, then give them the option of entering in a measurement from a tape measure.
This answer is somewhat similar to Jitendra's answer, but the method he suggests will only work where you can fit the whole wall in a single shot.
Get an object of know size and photograph it held against the wall with the iphone held against the other wall (two people or blutac needed). Then you can calculate the distance between the walls by looking at the size of the object (in pixels) in the photo. You could use a PDF to make a printed document the object of known size and use a 2D barcode to get the iphone to pick it up.
When the user wants to measure something, he takes a picture of it, but you're actually taking two separate images, one with flash on, one with flash off. Then you can analyze the lighting differences in the image and the flash reflection to determine the scale of the image. This will only work for close and not too shining objects I guess.
But that's about the only other way I thought about deducting scale from an image without any fixed objects.