I have a set of videos of someone talking, I'm building a lip recognition system therefore I need to perform some image processing on specific region of the image (the lower chin and lips).
I have over 200 videos, each containing a sentence. It is natural conversation therefore the head constantly moves so the lips aren't in a fixed place. I'm having difficulty specifying my region of interest in the image as it is very tiresome having to watch through every video and mark out how big my box to ensure the lips are cropped within the ROI.
I was wondering if there would be an easier way to check this, perhaps using MATLAB? I was thinking I could crop the video frame by frame and output an image for each frame. And then physically go through the images to see if the lips go out of frame?
I had to solve a similar problem dealing with tracking the heads and limbs of students participating in class discussions on video. We experimented with using state of the art optical flow tracking, from Thomas Brox ( link, see the part about large-displacement optical flow.) In our case, we had nearly 20 terabytes of videos to work through, so we had no choice but to use a C++ and GPU implementation of optical flow code; I think you will discover too that Matlab is impossibly slow for doing video analysis.
Optical flow returns to you detailed motion vectors. Then, if you can just mark the original bounding box for the mouth and chin in the first frame of the video, you can follow the tracks given by the optical flow of those pixels and this will usually give you a good sequence of bounding boxes. You will probably have errors that you have to clean up, though. You could write a Python script that plays back the sequence of bounding boxes for you to quickly check for errors though.
The code I wrote for this is in Python, and it's probably not easy to adapt to your data set-up or your problem, but you can find my affine-transformation based optical flow tracking code linked here, in the part called 'Object tracker using dense optical flow.'
The short answer is that this is a very difficult and annoying problem for vision researchers. Most people "solve" it by placing their videos, frame by frame, onto Mechanical Turk, and paying human workers about 2 cents per frame that they analyze. This gives you pretty good results (you'll still have to clean them after collecting it from the Mechanical Turkers), but it's not very helpful when you have tons o' videos and you cannot wait for enough of them to randomly get analyzed on Mechanical Turk.
There definitely isn't any 'out of the box' solution to region-of-interest annotation, though. You'd probably have to pay quite a lot for third-party software that did this automatically. My best guess for that is to check out what face.com would charge you and how well it would perform. Be careful that you don't violate any researcher confidentiality agreements with your data set though, for this or Mechanical Turk.
Related
I want to remove background and get deer as a foreground image.
This is my source image captured by trail camera:
This is what I want to get. This output image can be a binary image or RGB.
I worked on it and try many methods to get solution but every time it failed at specific point. So please first understand what is my exact problem.
Image are captured by a trail camera and camera is motion detector. when deer come in front of camera it capture image.
Scene mode change with respect to weather changing or day and night etc. So I can't use frame difference or some thing like this.
Segmentation may be not work correctly because Foreground (deer) and Background have same color in many cases.
If anyone still have any ambiguity in my question then please first ask me to clear and then answer, it will be appreciated.
Thanks in advance.
Here's what I would do:
As was commented to your question, you can detect the dear and then perform grabcut to segment it from the picture.
To detect the dear, I would couple a classifier with a sliding window approach. That would mean that you'll have a classifier that given a patch (can be a large patch) in the image, output's a score of how much that patch is similar to a dear. The sliding window approach means that you loop on the window size and then loop on the window location. For each position of the window in the image, you should apply the classifier on that window and get a score of how much that window "looks like" a dear. Once you've done that, threshold all the scores to get the "best windows", i.e. the windows that are most similar to a dear. The rational behind this is that if we a dear is present at some location in the image, the classifier will output a high score at all windows that are close/overlap with the actual dear location. We would like to merge all that locations to a single location. That can be done by applying the functions groupRectangles from OpenCV:
http://docs.opencv.org/modules/objdetect/doc/cascade_classification.html#grouprectangles
Take a look at some face detection example from OpenCV, it basically does the same (sliding window + classifier) where the classifier is a Haar cascade.
Now, I didn't mention what that "dear classifier" can be. You can use HOG+SVM (which are both included in OpenCV) or use a much powerful approach of running a deep convulutional neural network (deep CNN). Luckily, you don't need to train a deep CNN. You can use the following packages with their "off the shelf" ImageNet networks (which are very powerful and might even be able to identify a dear without further training):
Decaf- which can be used only for research purposes:
https://github.com/UCB-ICSI-Vision-Group/decaf-release/
Or Caffe - which is BSD licensed:
http://caffe.berkeleyvision.org/
There are other packages of which you can read about here:
http://deeplearning.net/software_links/
The most common ones are Theano, Cuda ConvNet's and OverFeat (but that's really opinion based, you should chose the best package from the list that I linked to).
The "off the shelf" ImageNet network were trained on roughly 10M images from 1000 categories. If those categories contain "dear", that you can just use them as is. If not, you can use them to extract features (as a 4096 dimensional vector in the case of Decaf) and train a classifier on positive and negative images to build a "dear classifier".
Now, once you detected the dear, meaning you have a bounding box around it, you can apply grabcut:
http://docs.opencv.org/trunk/doc/py_tutorials/py_imgproc/py_grabcut/py_grabcut.html
You'll need an initial scribble on the dear to perform grabcu. You can just take a horizontal line in the middle of the bounding box and hope that it will be on the dear's torso. More elaborate approaches would be to find the symmetry axis of the dear and use that as a scribble, but you would have to google, research an implement some method to extract symmetry axis from the image.
That's about it. Not straightforward, but so is the problem.
Please let me know if you have any questions.
Try OpenCV Background Substraction with Mixture of Gaussians models. They should be adaptable enough for your scenes. Of course, the final performance will depend on the scenario, but it is worth trying.
Since you just want to separate the background from the foreground I think you do not need to recognize the deer. You need to recognize an object in motion in the scene. You just need to separate what is static in a significant interval of time (background) from what is not static: the deer.
There are algorithms that combine multiple frames from the same scene in order to determine the background, like THIS ONE.
You mentioned that the scene mode changes with respect to weather changing or day and night considering photos of different deers.
You could implement a solution when motion is detected, instead of taking a single photo, it could take a few ones with some interval of time.
This interval has to be long as to get the deer in different positions or out of the scene and at the same time short enough to not be much affected by scene variations. Perhaps you need to deal with some brightness variation, but I think it is feasible to determine the background using these frames and finally segment the deer in the "motion frame".
I have a web cam that takes a picture every N seconds. This gives me a collection of images of the same scene over time. I want to process that collection of images as they are created to identify events like someone entering into the frame, or something else large happening. I will be comparing images that are adjacent in time and fixed in space - the same scene at different moments of time.
I want a reasonably sophisticated approach. For example, naive approaches fail for outdoor applications. If you count the number of pixels that change, for example, or the percentage of the picture that has a different color or grayscale value, that will give false positive reports every time the sun goes behind a cloud or the wind shakes a tree.
I want to be able to positively detect a truck parking in the scene, for example, while ignoring lighting changes from sun/cloud transitions, etc.
I've done a number of searches, and found a few survey papers (Radke et al, for example) but nothing that actually gives algorithms that I can put into a program I can write.
Use color spectroanalisys, without luminance: when the Sun goes down for a while, you will get similar result, colors does not change (too much).
Don't go for big changes, but quick changes. If the luminance of the image changes -10% during 10 min, it means the usual evening effect. But when the change is -5%, 0, +5% within seconds, its a quick change.
Don't forget to adjust the reference values.
Split the image to smaller regions. Then, when all the regions change same way, you know, it's a global change, like an eclypse or what, but if only one region's parameters are changing, then something happens there.
Use masks to create smart regions. If you're watching a street, filter out the sky, the trees (blown by wind), etc. You may set up different trigger values for different regions. The regions should overlap.
A special case of the region is the line. A line (a narrow region) contains less and more homogeneous pixels than a flat area. Mark, say, a green fence, it's easy to detect wheter someone crosses it, it makes bigger change in the line than in a flat area.
If you can, change the IRL world. Repaint the fence to a strange color to create a color spectrum, which can be identified easier. Paint tags to the floor and wall, which can be OCRed by the program, so you can detect wheter something hides it.
I believe you are looking for Template Matching
Also i would suggest you to look on to Open CV
We had to contend with many of these issues in our interactive installations. It's tough to not get false positives without being able to control some of your environment (sounds like you will have some degree of control). In the end we looked at combining some techniques and we created an open piece of software named OpenTSPS (Open Toolkit for Sensing People in Spaces - http://www.opentsps.com). You can look at the C++ source in github (https://github.com/labatrockwell/openTSPS/).
We use ‘progressive background relearn’ to adjust to the changing background over time. Progressive relearning is particularly useful in variable lighting conditions – e.g. if lighting in a space changes from day to night. This in combination with blob detection works pretty well and the only way we have found to improve is to use 3D cameras like the kinect which cast out IR and measure it.
There are other algorithms that might be relevant, like SURF (http://achuwilson.wordpress.com/2011/08/05/object-detection-using-surf-in-opencv-part-1/ and http://en.wikipedia.org/wiki/SURF) but I don't think it will help in your situation unless you know exactly the type of thing you are looking for in the image.
Sounds like a fun project. Best of luck.
The problem you are trying to solve is very interesting indeed!
I think that you would need to attack it in parts:
As you already pointed out, a sudden change in illumination can be problematic. This is an indicator that you probably need to achieve some sort of illumination-invariant representation of the images you are trying to analyze.
There are plenty of techniques lying around, one I have found very useful for illumination invariance (applied to face recognition) is DoG filtering (Difference of Gaussians)
The idea is that you first convert the image to gray-scale. Then you generate two blurred versions of this image by applying a gaussian filter, one a little bit more blurry than the first one. (you could use a 1.0 sigma and a 2.0 sigma in a gaussian filter respectively) Then you subtract from the less-blury image, the pixel intensities of the more-blurry image. This operation enhances edges and produces a similar image regardless of strong illumination intensity variations. These steps can be very easily performed using OpenCV (as others have stated). This technique has been applied and documented here.
This paper adds an extra step involving contrast equalization, In my experience this is only needed if you want to obtain "visible" images from the DoG operation (pixel values tend to be very low after the DoG filter and are veiwed as black rectangles onscreen), and performing a histogram equalization is an acceptable substitution if you want to be able to see the effect of the DoG filter.
Once you have illumination-invariant images you could focus on the detection part. If your problem can afford having a static camera that can be trained for a certain amount of time, then you could use a strategy similar to alarm motion detectors. Most of them work with an average thermal image - basically they record the average temperature of the "pixels" of a room view, and trigger an alarm when the heat signature varies greatly from one "frame" to the next. Here you wouldn't be working with temperatures, but with average, light-normalized pixel values. This would allow you to build up with time which areas of the image tend to have movement (e.g. the leaves of a tree in a windy environment), and which areas are fairly stable in the image. Then you could trigger an alarm when a large number of pixles already flagged as stable have a strong variation from one frame to the next one.
If you can't afford training your camera view, then I would suggest you take a look at the TLD tracker of Zdenek Kalal. His research is focused on object tracking with a single frame as training. You could probably use the semistatic view of the camera (with no foreign objects present) as a starting point for the tracker and flag a detection when the TLD tracker (a grid of points where local motion flow is estimated using the Lucas-Kanade algorithm) fails to track a large amount of gridpoints from one frame to the next. This scenario would probably allow even a panning camera to work as the algorithm is very resilient to motion disturbances.
Hope this pointers are of some help. Good Luck and enjoy the journey! =D
Use one of the standard measures like Mean Squared Error, for eg. to find out the difference between two consecutive images. If the MSE is beyond a certain threshold, you know that there is some motion.
Also read about Motion Estimation.
if you know that the image will remain reletivly static I would reccomend:
1) look into neural networks. you can use them to learn what defines someone within the image or what is a non-something in the image.
2) look into motion detection algorithms, they are used all over the place.
3) is you camera capable of thermal imaging? if so it may be worthwile to look for hotspots in the images. There may be existing algorithms to turn your webcam into a thermal imager.
Let's say you are taking a video (with the camera in a steady position) and a bird flies through the view of the camera. It should be possible to do image segmentation and automatically remove this bird from the video.
What are these styles of algorithms called and how are they normally accomplished?
There's a technique called Simple Image Object Extraction (SIOX) - it uses a technique to identify foreground vs. background objects in still and video images. The open source GIMP editor has an implementation of it, and there's more information about it here.
From the overview:
SIOX stands for Simple Interactive Object Extraction and is a solution for extracting foreground from still images with very little user interaction. SIOX is fast, noise robust, and can therefore also be used for the segmentation of videos. It avoids many of the drawbacks of graph-based segmentation methods but performs about equally well on different benchmarks. SIOX is open and free (Apache License) and the authors have intentionally not patented any part of the technology. As a result, it has been integrated into several open-source image manipulation programs over the past years. SIOX is the underlying algorithm of the foreground extraction tool in the GNU Image Manipulation Program (GIMP) and is part of the tracer tool in Inkscape. SIOX originates from E-Chalk where an instructor standing in front of an electronic chalkboard is segmented. Variants of SIOX are being used for robotic vision and for improving 3D time-of-flight camera segmentation.
Here's a link to the Java Reference Implementation of SIOX.
Here's a link to the PDF with details about how a variation of the algorithm works.
You should be able to adapt it to use inter-frame interpolation to remove a specific foreground object from each frame of a video by using temporal data from surrounding frames.
If the camera is fixed and there isn't too much motion in the scene, then I would suggest a method based on background subtraction.
Step 1: Compute background for each frame of the video. There are complicated algorithms for doing this, but a very simple and effective one would be to compute the median value of every pixel in the image across a 3 second time window. Longer if the object in question is moving slowly. Incidentally, if you just perform this kind of filtering it will remove most moving objects from the video if the camera is fixed, hence my earlier question about all objects vs. one object.
Step 2: Mark the regions you want to remove in each frame with a brush tool, and replace them with the background pixels. Don't bother with a fine brush or lasso tool as any non-object pixels you mark will just be replaced with their filtered version. You could probably use the same brush marks for several frames since the boundary is not so important. If the object is the only thing moving in the scene, you could just mark the entire frame and have it replaced with the background.
Anyways, to answer your more general question, the topic you want to research is called inpainting for images and video. There is quite a bit of literature out there on the subject, what I described was just a super simple method you could implement in an hour or so with opencv.
After asking here and trying both SURF and SIFT, neither of them seams to be efficient enough to generate interest points fast enough to track a stream from the camera.
SURF, for example, takes around 3 seconds to generate interest points for an image, that's way too slow to track a video coming from a web cam, and it'll be even worse when using it on a mobile phone.
I just need an algorithm that tracks a certain area, its scale, tilt, etc.. and I can build on top of that.
Thanks
I suspect your SURF usage may need some alteration?
Here is a link to an MIT paper on using SURF for augmented reality applications on mobile devices.
Excerpt:
In this section, we present our
implementation of the SURF al- gorithm
and its adaptation to the mobile
phone. Next, we discuss the impact
that accuracy has on the speed of the
nearest-neighbor search and show that
we can achieve an order of magnitude
speed- up with minimal impact on
matching accuracy. Finally, we dis-
cuss the details of the phone
implementation of the image matching
pipeline. We study the performance,
memory use, and bandwidth consumption
on the phone.
You might also want to look into OpenCV's algorithms because they are tried and tested.
Depending on the constraints of your application, you may be able to reduce the genericness of those algorithms to look for known POIs and markers within the image.
Part of tracking a POI is estimating its vector from one point in the 2D image to another, and then optionally confirming that it still exists there (through pixel characteristics). The same approach can be used to track (not re-scan the entire image) for POI and POI group/object perspective and rotation changes.
There are tons of papers online for tracking objects on a 2D projection (up to a servere skew in many cases).
Good Luck!
You should try FAST detector
http://svr-www.eng.cam.ac.uk/~er258/work/fast.html
We are using SURF for a project and we found OpenSURF to outmatch OpenCV's SURF implementation in raw speed and performance. We still haven´t tested repeatability and accuracy, but it is way faster.
Update:
I just wanted to point out that you needn't perform a SURF match step in each frame, you could simply do it every other frame and interpolate the position of the object in the frame you don't execute SURF on.
You can use a simpler algorithm if you would make stricter restrictions on the area you would like to be tracked. As you surely know, ARToolKit is pretty fast, but only tracks black and white markers with a very distinct frame.
If you want a (somewhat) general purpose tracker, you may want to check PTAM. The site (http://www.robots.ox.ac.uk/~gk/PTAM/) is currently down, but here's a snazzy video of it working on an iPhone (http://www.youtube.com/watch?v=pBI5HwitBX4)
As others have mentioned, three seconds seems unusually long. While testing the SURF implementation in the Mahotas library, I found that it took on average 0.36sec, even with some fairly large images (e.g. 1024x768). And that's with a mix of Python and C, so I'd imagine some other pure-C implementations would be even faster.
I found this nice comparison of each feature detection algorithms at http://computer-vision-talks.com/2011/01/comparison-of-the-opencvs-feature-detection-algorithms-2/
Have a look. It might be useful!
According to that comparison, and as mirror2image has also suggested, FAST is the best choice. But it depends on what you really want to achieve.
One option I've used in constrained embedded systems is to use a simpler interest point detector: FAST or Shi-Tomasi for example. I used Shi-Tomasi, as I was targetting an FPGA and could easily run it at pixel rate with no significant buffering required.
Then use SURF to generate the descriptors for the image patch around the identified features and use those for matching and tracking purposes.
I am building a web application using .NET 3.5 (ASP.NET, SQL Server, C#, WCF, WF, etc) and I have run into a major design dilemma. This is a uni project btw, but it is 100% up to me what I develop.
I need to design a system whereby I can take an image and automatically crop a certain object within it, without user input. So for example, cut out the car in a picture of a road. I've given this a lot of thought, and I can't see any feasible method. I guess this thread is to discuss the issues and feasibility of achieving this goal. Eventually, I would get the dimensions of a car (or whatever it may be), and then pass this into a 3d modelling app (custom) as parameters, to render a 3d model. This last step is a lot more feasible. It's the cropping issue which is an issue. I have thought of all sorts of ideas, like getting the colour of the car and then the outline around that colour. So if the car (example) is yellow, when there is a yellow pixel in the image, trace around it. But this would fail if there are two yellow cars in a photo.
Ideally, I would like the system to be completely automated. But I guess I can't have everything my way. Also, my skills are in what I mentioned above (.NET 3.5, SQL Server, AJAX, web design) as opposed to C++ but I would be open to any solution just to see the feasibility.
I also found this patent: US Patent 7034848 - System and method for automatically cropping graphical images
Thanks
This is one of the problems that needed to be solved to finish the DARPA Grand Challenge. Google video has a great presentation by the project lead from the winning team, where he talks about how they went about their solution, and how some of the other teams approached it. The relevant portion starts around 19:30 of the video, but it's a great talk, and the whole thing is worth a watch. Hopefully it gives you a good starting point for solving your problem.
What you are talking about is an open research problem, or even several research problems. One way to tackle this, is by image segmentation. If you can safely assume that there is one object of interest in the image, you can try a figure-ground segmentation algorithm. There are many such algorithms, and none of them are perfect. They usually output a segmentation mask: a binary image where the figure is white and the background is black. You would then find the bounding box of the figure, and use it to crop. The thing to remember is that none of the existing segmentation algorithm will give you what you want 100% of the time.
Alternatively, if you know ahead of time what specific type of object you need to crop (car, person, motorcycle), then you can try an object detection algorithm. Once again, there are many, and none of them are perfect either. On the other hand, some of them may work better than segmentation if your object of interest is on very cluttered background.
To summarize, if you wish to pursue this, you would have to read a fair number of computer vision papers, and try a fair number of different algorithms. You will also increase your chances of success if you constrain your problem domain as much as possible: for example restrict yourself to a small number of object categories, assume there is only one object of interest in an image, or restrict yourself to a certain type of scenes (nature, sea, etc.). Also keep in mind, that even the accuracy of state-of-the-art approaches to solving this type of problems has a lot of room for improvement.
And by the way, the choice of language or platform for this project is by far the least difficult part.
A method often used for face detection in images is through the use of a Haar classifier cascade. A classifier cascade can be trained to detect any objects, not just faces, but the ability of the classifier is highly dependent on the quality of the training data.
This paper by Viola and Jones explains how it works and how it can be optimised.
Although it is C++ you might want to take a look at the image processing libraries provided by the OpenCV project which include code to both train and use Haar cascades. You will need a set of car and non-car images to train a system!
Some of the best attempts I've see of this is using a large database of images to help understand the image you have. These days you have flickr, which is not only a giant corpus of images, but it's also tagged with meta-information about what the image is.
Some projects that do this are documented here:
http://blogs.zdnet.com/emergingtech/?p=629
Start with analyzing the images yourself. That way you can formulate the criteria on which to match the car. And you get to define what you cannot match.
If all cars have the same background, for example, it need not be that complex. But your example states a car on a street. There may be parked cars. Should they be recognized?
If you have access to MatLab, you could test your pattern recognition filters with specialized software like PRTools.
Wwhen I was studying (a long time ago:) I used Khoros Cantata and found that an edge filter can simplify the image greatly.
But again, first define the conditions on the input. If you don't do that you will not succeed because pattern recognition is really hard (think about how long it took to crack captcha's)
I did say photo, so this could be a black car with a black background. I did think of specifying the colour of the object, and then when that colour is found, trace around it (high level explanation). But, with a black object in a black background (no constrast in other words), it would be a very difficult task.
Better still, I've come across several sites with 3d models of cars. I could always use this, stick it into a 3d model, and render it.
A 3D model would be easier to work with, a real world photo much harder. It does suck :(
If I'm reading this right... This is where AI shines.
I think the "simplest" solution would be to use a neural-network based image recognition algorithm. Unless you know that the car will look the exact same in each picture, then that's pretty much the only way.
If it IS the exact same, then you can just search for the pixel pattern, and get the bounding rectangle, and just set the image border to the inner boundary of the rectangle.
I think that you will never get good results without a real user telling the program what to do. Think of it this way: how should your program decide when there is more than 1 interesting object present (for example: 2 cars)? what if the object you want is actually the mountain in the background? what if nothing of interest is inside the picture, thus nothing to select as the object to crop out? etc, etc...
With that said, if you can make assumptions like: only 1 object will be present, then you can have a go with using image recognition algorithms.
Now that I think of it. I recently got a lecture about artificial intelligence in robots and in robotic research techniques. Their research went on about language interaction, evolution, and language recognition. But in order to do that they also needed some simple image recognition algorithms to process the perceived environment. One of the tricks they used was to make a 3D plot of the image where x and y where the normal x and y axis and the z axis was the brightness of that particular point, then they used the same technique for red-green values, and blue-yellow. And lo and behold they had something (relatively) easy they could use to pick out the objects from the perceived environment.
(I'm terribly sorry, but I can't find a link to the nice charts they had that showed how it all worked).
Anyway, the point is that they were not interested (that much) in image recognition so they created something that worked good enough and used something less advanced and thus less time consuming, so it is possible to create something simple for this complex task.
Also any good image editing program has some kind of magic wand that will select, with the right amount of tweaking, the object of interest you point it on, maybe it's worth your time to look into that as well.
So, it basically will mean that you:
have to make some assumptions, otherwise it will fail terribly
will probably best be served with techniques from AI, and more specifically image recognition
can take a look at paint.NET and their algorithm for their magic wand
try to use the fact that a good photo will have the object of interest somewhere in the middle of the image
.. but i'm not saying that this is the solution for your problem, maybe something simpler can be used.
Oh, and I will continue to look for those links, they hold some really valuable information about this topic, but I can't promise anything.