Coding4Fun Articles

MJPEG Decoder


Last year the Coding4Fun/Channel 9 guys asked me to work on a few things for MIX10.  One of these items was a way to output a webcam stream to Windows Phone 7 for use with Clint's t-shirt cannon project you may have read about.  I figured the easiest way to accomplish this was by using a network/IP camera capable of sending a Motion-JPEG stream, which can be easily decoded and displayed that can display a JPEG image.  Thus, this library was born.

It has gone through quite a few changes and I have expanded it to easily display MJPEG streams on a variety of platforms.  The developer just references the assembly appropriate to their platform, adds a few lines of code, and away it goes.


For those that are just interested in the usage, it's as simple as this:

  1. Reference one of the following assemblies appropriate for your project:
    • MjpegProcessorSL.dll - Silverlight (Out of Browser Only!)
    • MjpegProcessorWP7.dll - Windows Phone 7 (XNA or Silverlight, performance maxes out around 320x240 @ 15fps, so set your camera settings accordingly)
    • MjpegProcessorXna4.dll - XNA 4.0 (Windows)
    • MjpegProcessor.dll - WinForms and WPF
  2. Create a new MjpegDecoder object.
  3. Hook up the FrameReady event.
  4. In the event handler, take the Bitmap/BitmapImage and assign it to your image display control:
  5. In the case of XNA, use the GetMjpegFrame method in the Update method, which will return a Texture2D you can use in your Draw method.
  6. Call the ParseStream method with the Uri of the MJPEG "endpoint".

That's it!  The source code and binaries above both include projects demonstrating how to use the library on each of these platforms.  As long as you set the appropriate reference, you can just copy and paste the code in the sample to get your project running (changing the Uri, of course).

public partial class MainWindow : Window
    MjpegDecoder _mjpeg;

    public MainWindow()
        _mjpeg = new MjpegDecoder();
        _mjpeg.FrameReady += mjpeg_FrameReady;

    private void Start_Click(object sender, RoutedEventArgs e)
        _mjpeg.ParseStream(new Uri(""));

    private void mjpeg_FrameReady(object sender, FrameReadyEventArgs e)
        image.Source = e.BitmapImage;

If that doesn't fit your needs, you can also access the Bitmap/BitmapImage properties directly from the MjpegDecoder object, or the CurrentFrame property, which will contain the raw JPEG data prior to being decoded.

A Word About Network/IP Cameras

I have tested this against several different cameras.  Each device has its own quirks, but all of them seem to work with this library with one exception:  several cameras will respond differently when an Internet Explorer user agent header is sent with the HTTP request.  Instead of sending down an MJPEG stream, it will send a single JPEG image as Internet Explorer does not properly support MJPEG streams.  Unfortunately, this causes the Silverlight processor to not work properly as the header cannot be changed from the Internet Explorer default.  When this happens, only a single frame will be sent, and the decoding will fail.  The only fix I have found is to use a different camera that doesn't work in this way.

What is MJPEG?

Pretty simply, it's a video format where each frame of video is sent as a separate, compressed JPEG image.  A standard HTTP request is made to a specific URL, and a multipart response is sent.  Parsing this multipart stream into separate images as they are sent results in a series of JPEG images.  The viewer displays those JPEG images as quickly as they are sent and that creates the video.  It's not a well documented format, nor is it perfectly standardized, but it does work.  For more information, see the MJPEG article on Wikipedia.

How Do I Find the MJPEG URL of My Camera?

Excellent question.  Not an excellent answer.  The user manual may mention the URL.  A quick internet search with the model number should get you a result.  Or, you can also try this company's lookup tool.

How Does It Work?

Glad you asked.  If you take a look at the project, you'll notice there isn't much code.  One single file is used with a variety of compiler directives to compile certain portions based on the platform assembly being generated.  The MjpegDecoder.cs/.vb contains the entire implementation.

First, an asynchronous request is made to the provided MJPEG URL inside the ParseStream method.  If we are in a Silverlight environment, the AllowReadStreamBuffering property must be set to false so the response returns immediately instead of being buffered.  Additionally, we need to register the http:// prefix to use the client http stack vs. the browser stack.  Finally, the request is made using the BeginGetResponse method, specifying the OnGetResponse method as the callback.  This will be called as soon as data is sent from the camera in response to our request.

public void ParseStream(Uri uri)
    HttpWebRequest.RegisterPrefix("http://", WebRequestCreator.ClientHttp);
    HttpWebRequest request = (HttpWebRequest)HttpWebRequest.Create(uri);

    // start the stream immediately
    request.AllowReadStreamBuffering = false;
    // asynchronously get a response
    request.BeginGetResponse(OnGetResponse, request);

OnGetResponse grabs the response headers and uses the Content-Type header to determine the boundary marker that will be sent between JPEG frames.

private void OnGetResponse(IAsyncResult asyncResult)
    HttpWebResponse resp;
    byte[] buff;
    byte[] imageBuffer = new byte[1024 * 1024];
    Stream s;

    // get the response
    HttpWebRequest req = (HttpWebRequest)asyncResult.AsyncState;
    resp = (HttpWebResponse)req.EndGetResponse(asyncResult);

    // find our magic boundary value
    string contentType = resp.Headers["Content-Type"];
    if(!string.IsNullOrEmpty(contentType) && !contentType.Contains("="))
        throw new Exception("Invalid content-type header.  The camera is likely not returning a proper MJPEG stream.");
    string boundary = resp.Headers["Content-Type"].Split('=')[1];
    byte[] boundaryBytes = Encoding.UTF8.GetBytes(boundary.StartsWith("--") ? boundary : "--" + boundary);


It then streams the response data, looks for a JPEG header marker, then reads until it finds the boundary marker, copies the data into a buffer, decodes it, passes it on to whoever wants it via an event, and then starts over.

    s = resp.GetResponseStream();
    BinaryReader br = new BinaryReader(s);

    _streamActive = true;

    buff = br.ReadBytes(ChunkSize);

    while (_streamActive)
        int size;

        // find the JPEG header
        int imageStart = buff.Find(JpegHeader);

        if(imageStart != -1)
            // copy the start of the JPEG image to the imageBuffer
            size = buff.Length - imageStart;
            Array.Copy(buff, imageStart, imageBuffer, 0, size);

                buff = br.ReadBytes(ChunkSize);

                // find the boundary text
                int imageEnd = buff.Find(boundaryBytes);
                if(imageEnd != -1)
                    // copy the remainder of the JPEG to the imageBuffer
                    Array.Copy(buff, 0, imageBuffer, size, imageEnd);
                    size += imageEnd;

                    // create a single JPEG frame
                    CurrentFrame = new byte[size];
                    Array.Copy(imageBuffer, 0, CurrentFrame, 0, size);
#if !XNA
                    // copy the leftover data to the start
                    Array.Copy(buff, imageEnd, buff, 0, buff.Length - imageEnd);

                    // fill the remainder of the buffer with new data and start over
                    byte[] temp = br.ReadBytes(imageEnd);

                    Array.Copy(temp, 0, buff, buff.Length - imageEnd, temp.Length);

                // copy all of the data to the imageBuffer
                Array.Copy(buff, 0, imageBuffer, size, buff.Length);
                size += buff.Length;

The ProcessFrame method seen above takes the raw byte buffer, which contains an undecoded JPEG image, and decodes it based on the environment.  However this isn't called in the case of XNA which we'll see in a moment:

private void ProcessFrame(byte[] frameBuffer)
    // need to get this back on the UI thread
    Deployment.Current.Dispatcher.BeginInvoke((Action)(() =>
        // resets the BitmapImage to the new frame
        BitmapImage.SetSource(new MemoryStream(frameBuffer, 0, frameBuffer.Length));

        // tell whoever's listening that we have a frame to draw
        if(FrameReady != null)
            FrameReady(this, new FrameReadyEventArgs { FrameBuffer = CurrentFrame, BitmapImage = BitmapImage });

    // I assume if there's an Application.Current then we're in WPF, not WinForms
    if(Application.Current != null)
        // get it on the UI thread
        Application.Current.Dispatcher.BeginInvoke((Action)(() =>
            // create a new BitmapImage from the JPEG bytes
            BitmapImage = new BitmapImage();
            BitmapImage.StreamSource = new MemoryStream(frameBuffer);

            // tell whoever's listening that we have a frame to draw
            if(FrameReady != null)
                FrameReady(this, new FrameReadyEventArgs { FrameBuffer = CurrentFrame, Bitmap = Bitmap, BitmapImage = BitmapImage });
        // create a simple GDI+ happy Bitmap
        Bitmap = new Bitmap(new MemoryStream(frameBuffer));

        // tell whoever's listening that we have a frame to draw
        if(FrameReady != null)
            FrameReady(this, new FrameReadyEventArgs { FrameBuffer = CurrentFrame, Bitmap = Bitmap, BitmapImage = BitmapImage });

In the case of Silverlight, the BitmapImage object has a SetSource method which takes a stream to be decoded and turned into the image.  In WPF, BitmapImage works differently.  In this case, BeginInit is called, then the StreamSource property is set to the stream of bytes, and finally EndInit is called.  In WinForms, the library will return a Bitmap object which can be initialized with the stream right in the constructor.

In the code above, I look at the Application.Current property to determine if the library is being used by a WPF project.  If that property is not null, it is assumed the library is being called from a WPF project.

When compiled as an XNA library, we have no use for a BitmapImage or a Bitmap…we need a Texture2D object.  The GetMjpegFrame method seen below is what is called by an XNA application during the Update method to pull the current frame:

public Texture2D GetMjpegFrame(GraphicsDevice graphicsDevice)
    // create a Texture2D from the current byte buffer
    if(CurrentFrame != null)
        return Texture2D.FromStream(graphicsDevice, new MemoryStream(CurrentFrame, 0, CurrentFrame.Length));
    return null;


And that's about it for the library.  Give it a try and please contact me with feedback or if you run into any issues.  Enjoy!

The Discussion

  • LukePuplett

    Without recommending any manufacturer in particular, can you recommend a particular manufacturer of camera we should use for this? Wink

    ...ideally I'd need it to be waterproof and almost invisible when placed behind bathroom tiles.


  • BrianPeek

    @LukePuplett:For your scenario, I'd recommend any of the cameras manufactured by Pervsys.  Big Smile

  • LukePuplett

    Thanks Brian, sounds like my kind of brand.

  • stuat

    can you do this using MediaStreamSource and the raw av pipeline in silverlight ??

  • _ivan

    @BrianPeek: can you say which camera(s) use used in this project? Or can you recommend standard(not for LukePuplett use case) cameras which have MJPEG streaming easily accessible?

  • _ivan

    By the way thanks for the library, it is very timely, I hope it will be a good replacement for what I currently use.

  • BrianPeek

    @stuat: I started to write a MediaStreamSource implementation but didn't bother to finish.  You should be able to shove this into a MediaStreamSource without too much issue, though.

  • BrianPeek

    @_ivan: I have tested with the Cisco WVC210, TRENDnet IP110W, and a very old no-name brand camera.  Practically any IP/network camera (i.e. not a USB webcam) should output MJPEG and work with this.  I don't think I've seen an IP cam without some kind of MJPEG support, but that doesn't mean they don't exist.

  • barrkel

    Why not just use TcpClient instead of HttpWebRequest to get around the limitation on the IE User-Agent header? It's not like it's difficult to write a HTTP 1.0 client, especially if it only needs to GET a single resource.

    But then I see TcpClient isn't supported in Silverlight. That really sucks.

  • LukePuplett

    So, do out-of-browser apps have that HTTP header limitation?

  • BrianPeek

    @LukePuplett: You can only run this OOB becuase of the crossdomain.xml policy goodness in Silverlight.  The cameras don't have this file on their internal webserver, so browser-based Silverlight won't work at all.  That said, even OOB apps can't modify the User-Agent header either via the Headers collection or the UserAgent property.

  • Nils

    @barrkel: I'm note quite sure but Silverlight 4 should be able to use System.Net.Sockets.Socket to connect to any resource when running with Elevated Trust as described at

    But this looks like a very low-level support of sockets and I think it requires some work to implement a full-featured TCP Client on top of the Socket class.

    Besides that, nice work, Brian.

    (Just my 2 cents...)

  • _ivan

    @BrianPeek: oh good, I used TRENDNet IP cameras in the past for these kind of tasks, and was very pleased with their simplicity.

  • BrianPeek
    @nils: Definitely an option worth exploring. Anybody want to give it a try? I'm swamped for the the next few weeks on some other projects. Smiley. I'd be happy to add it in if anyone manages to make it work...
  • Magdur

    nice work, Brian



  • Bert Vissers

    Very nice article Brian, keep up the good work!

  • Bojan

    I am using your MJPEG Decoder for display JPEG from our server. Owner of the server is sayng that we have produce heavy load on the server. Is this possible?
    I am 99% soure that my question is stupid, but I must ask: is it possible that while loop which read bytes from BinaryReader, causes too much load on the server?!

  • BrianPeek

    @Bojan: what kind of load are you seeing and what version of the decoder are you using? (wpf, xna, etc.)  In my testing here, my test apps are cranking away at a 640x480 stream at 30fps and I'm only seeing 5-6% CPU at most, typically less.

  • Bojan

    I don't know how much is loading(CPU) exactly, I don't have access to the server. I am using decoder (for WinForms) for display streams from 20 cameras. I made my own server wich send multipart response with jpegs, so i wil test on him.

  • Bojan

    I forgot, jpeg dimension is 266x150 and framerate is 1000 (1 fps). According to your parameters, my parameters shouldn't cause problems on server?

  • BrianPeek

    @Bojan: Well, if you're constantly reading the stream of 20 cameras simultaneously, there's a chance it could be putting undue load on the server.  It has never been tested in that kind of environment.  I wouldn't think you'd crush the server, but if each stream is taking 2-3% of CPU to read/decode, make it times 20 and you're taking a decent toll on the server.

  • Reza Jafari

    it sounds fun

  • zipswich

    Thanks a lot, Brian, for this very informative article. I used WebClient instead of HttpWebRequest to make an SL app after reading your article.  I found it to be a little bit easier.  It works well.

    The SL x-domain access limitation almost defeats the purpose of the app which I made because of being tired of dealing with obsolete ActiveX controls one would have to install on every computer used to view an IP camera. I do not understand why SL does not allow in-browser X-domain access if it is allowed explicitly by the user for a specific uri.

    I am not sure what benefits one would get by using System.Net.Sockets as suggested by Nils.


  • Alex

    Do you know the AForge lib?

  • Misha

    hello! ssory, where it is possible to find the reference on MjpegProcessor.dll?

  • JimBeam

    Great library, great examples. However, when switching between two cameras (programatically, of course), the image object wants to show both streams at the same time. I have used the stopstream method but that doesn't seem to do anything and I can see both live feeds interlaced within each other. Help

  • Rene Tuchscherer

    Hi Brian,

    i use your great library in a windows phone 7 app.
    One question, how can i stop the stream when i leave the page there is a error (StopStream() don't work).

    Thx for René

  • Joseph Piron

    Hi Brian, thanks for this ! Really helpfull.
    But I have a small weird issue: I can't start more than two decoder in an application (I'm trying to get on stream from the same camera, can it be the problem ?) and if I stop the stream, it can't restart it ? Could it be related to a keep-alive or not closed connection ?

    Thanks !

Comments closed

Comments have been closed since this content was published more than 30 days ago, but if you'd like to continue the conversation, please create a new thread in our Forums, or Contact Us and let us know.