I\'ve been developing a face recognition application using EmguCV (C#). I got the whole thing working okay if I store the face images (training set) in simple windows folder
I couldnt make it work from reading a direct Stream from the the Database where the images are located but your workaround, saving the images to a local folder, worked for me, thx a lot for sharing.Here's my demo page where you load files from DB: http://www.edatasoluciones.com/FaceDetection/FaceDataBase
Finally made it!! just one more day of coding helped me to got the problem solved:
public void ProcessRequest(HttpContext context)
{
_httpContext = context;
var imageid = context.Request.QueryString["Image"];
if (imageid == null || imageid == "")
{
imageid = "1";
}
using (WebClient wc = new WebClient())
{
// Handler retrieves the image from database and load it on the stream
using (Stream s = wc.OpenRead("http://mypageurl/Image.ashx?Image=" + imageid))
{
using (Bitmap bmp = new Bitmap(s))
{
AddFace(bmp);
}
}
}
}
public void AddFace(Bitmap image)
{
var faceImage = DetectFace(image);
if (faceImage != null)
{
var stream = new MemoryStream();
faceImage.Save(stream, ImageFormat.Bmp);
stream.Position = 0;
byte[] data = new byte[stream.Length];
stream.Read(data, 0, (int)stream.Length);
_httpContext.Response.Clear();
_httpContext.Response.ContentType = "image/jpeg";
_httpContext.Response.BinaryWrite(data);
}
}
private Bitmap DetectFace(Bitmap faceImage)
{
var image = new Image<Bgr, byte>(faceImage);
var gray = image.Convert<Gray, Byte>();
string filePath = HttpContext.Current.Server.MapPath("haarcascade_frontalface_default.xml");
var face = new HaarCascade(filePath);
MCvAvgComp[][] facesDetected = gray.DetectHaarCascade(face, 1.1, 10, HAAR_DETECTION_TYPE.DO_CANNY_PRUNING, new Size(20, 20));
Image<Gray, byte> result = null;
foreach (MCvAvgComp f in facesDetected[0])
{
//draw the face detected in the 0th (gray) channel with blue color
image.Draw(f.rect, new Bgr(Color.Blue), 2);
result = image.Copy(f.rect).Convert<Gray, byte>();
break;
}
if (result != null)
{
result = result.Resize(200, 200, INTER.CV_INTER_CUBIC);
return result.Bitmap;
}
return null;
}
public bool IsReusable
{
get { return false; }
}
Check your fetchedBytes
array to see if you are consistently getting just a stream of bytes representing a BMP image (starting with 0x42 0x4D
), or if there may be "other stuff" in there, too.
Depending on how the BMP data was inserted into the Access database it may contain an OLE "wrapper". For example, an 8x8 24-bit BMP image of pure red is saved by MSPAINT.EXE like this
If I copy that file and paste it into a Bound Object Frame in an Access form then Access wraps the BMP data in some "OLE stuff" before writing it to the table. Later, if I try to retrieve the BMP image via code, using something like this...
Sub oleDumpTest()
Dim rst As ADODB.Recordset, ads As ADODB.Stream
Set rst = New ADODB.Recordset
rst.Open "SELECT * FROM TrainingSet1 WHERE ID = 1", Application.CurrentProject.Connection
Set ads = New ADODB.Stream
ads.Type = adTypeBinary
ads.Open
ads.Write rst("FaceImage").Value
rst.Close
Set rst = Nothing
ads.SaveToFile "C:\Users\Gord\Pictures\oleDump_red."
ads.Close
Set ads = Nothing
End Sub
...then the resulting file also contains the OLE "wrapper"...
...and obviously is not a valid stand-alone BMP file. If I rename that file to give it a .bmp
extension and try to open it in Paint, I get
So maybe (some of) the [FaceImage] objects in your database are not raw BMP data, and perhaps the other software is rejecting them (or simply not able to understand them).
Another possible issue is that when you get the images from files in a folder you hand the Image
object a string containing the file path...
trainingImages.Add(new Image<Gray, byte>(Application.StartupPath + "\\TrainedFaces\\" + LoadFaces));
...but when you try to retrieve the images from the database you hand the same object a Bitmap
object
MemoryStream stream = new MemoryStream(fetchedBytes);
bmpImage = new Bitmap(stream);
trainingImages.Add(new Emgu.CV.Image<Gray, Byte>(bmpImage));
I have no way of knowing whether the Emgu.CV.Image
object might behave differently depending on the type of object it is given, but a quick+dirty workaround might be to write bmpImage
to a temporary file, hand trainingImages.Add
the path to that file, and then delete the file.