Spark's profile - activity

Have a look at these blogs, which is really useful for stereo calibration.

http://blog.martinperis.com/2011/08/s...

http://blog.martinperis.com/2011/01/o...

Number of images and orientation of chess board being captured matters a lot. It is recommended to capture as many as 30 to 40 images in different orientation and different distance to make calibration best.

Yes, you are exactly right. This is due to auto gain and auto exposure property of camera. Try disabling AEC and AEC using some camera viewer application like amcap (windows) or cheese (linux) and try experimenting with different manual gain and exposure values suitable for your environment. Eager to see your results:-)

You can try out k-means segmentation. Here are few links,

http://www.pyimagesearch.com/2014/05/...

http://www.bogotobogo.com/python/Open...

http://stackoverflow.com/questions/20...

https://github.com/Itseez/opencv/blob...

Here are few supporting links for your question.

http://xanthippi.ceid.upatras.gr/peop...

http://www.jera.com/jbrewer/2014/01/c... - But with iOS

http://www.learncomputervision.com/ar...

Here are few links that would help you to start your work.

http://stackoverflow.com/questions/19...

http://www.youtube.com/watch?v=OMrgLp...

https://chrisjmccormick.wordpress.com...

http://www.magicandlove.com/blog/2011...

Yes, as kbarni said, this is expected due to the behavior of CCD and CMOS sensors. There will be considerable amount of noise being added to image (even though you don't wish for).

Here is the synthesized color sample shape image

And their hue channel image

In this synth image, there wont be any patches. Hope you are clear now.

It would be better to understand your issue if you post some sample images and results.

@sandun: Please read http://answers.opencv.org/faq/ before posting your questions. It looks like you dint get any output as you did not specified any argument to your application. argv is empty in your case, so imread fails and hence condition of src.empty satisfies. Put some messages in your code to debug where it break out.

Try using Amcap and see whether you are able to switch resolution with that application. If that works, then resolution switching should work with OpenCV as well. Link to download http://amcap-webcam-capture.en.lo4d.com/

Following code snippet would help to load raw image to OpenCV Mat structure and then you can do further processing like demosaicing or some detection algorithms.

Mat img;
FILE *fp = NULL;
char *imagedata = NULL;
int framesize = IMAGE_WIDTH * IMAGE_HEIGHT;

//Open raw Bayer image.
fp = fopen("filename.raw", "rb");

//Memory allocation for bayer image data buffer.
imagedata = (char*) malloc (sizeof(char) * framesize);

//Read image data and store in buffer.
fread(imagedata, sizeof(char), framesize, fp);

//Image dimension.
imageSize.height = IMAGE_WIDTH;
imageSize.width = IMAGE_HEIGHT;

//Create Opencv mat structure for image dimension. For 8 bit bayer, type should be CV_8UC1.
img.create(IMAGE_HEIGHT, IMAGE_WIDTH, CV_8UC1);

memcpy(img.data, imagedata, framesize);

free(imagedata);

fclose(fp);

//Perform demosaicing process
cvtColor(img, RGBImage, CV_BayerBG2BGR);

Once color space is converted to BGR, further processing like edge detection or enhancement can be done. Hope this helps.

Preprocessing of underwater images is explained in this paper - https://www.ensta-bretagne.fr/jaulin/paper_bazeille_CMM06.pdf

@zms: Share your input image as well.

Keep breakpoints in Visual studio and find which function creates assertion. Do some debugging on variable and its value. Then post your questions, so that others will know you have worked something and really need help.

@creative, mark answers if it was helpful

@KansaiRobot, mark answers if it was helpful

@ganesh17, mark answers if it was helpful

As berak said, it is always 24 BPP format. It can either be 8-bit gray scale with padded 16 bit or can also be BGR color data. Depends on data stored in the video.

You can refer to OpenCV samples. Many samples use this such concept. Here is one from \opencv\sources\samples\cpp\stereo_calib.cpp.

1. Create a text file with list of image names along with path to be processed.

2. Following code opens the text file and prepares the list of file names.

   FILE* f = fopen("ImageList.txt", "r");
   vector<string> imagelist;
   for(int i = 0; ; i++)
   {
   if(!fgets(buf, sizeof(buf)-3, f))
       break;
   size_t len = strlen(buf);
   
   while(len > 0 && isspace(buf[len-1]))
       buf[--len] = '\0';
   
   if(buf[0] == '#')
       continue;
   
   imagelist.push_back(buf);
   }
   fclose(f);
   

3. Start accessing images by imread(imagelist[i])

Assuming you have two curved lines running parallel to each other in your image, here is a suggestion to get the mid point between curved line.

1. Threshold the image
2. Find contour and its bounding rectangle for each curved line's contour.
3. Get angle of its bounding rectangle.
4. Bounding rectangle of both curved line must be same as it is parallel to each other. If the angle of bounding rect is 0 deg or 90 deg, subtract x coordinate of contour to get the mid point between parallel lines. If the angle of bounding rect is other than 0 deg or 90 deg, subtract y coordinate of contour.

Share your results with image samples.

@juanmanpr: Shape context in OpenCV 3.0 works well with their samples. But produces poor result with image attached in my original post. The distance value calculated by Shape context is high for same type of object, but low for different object, and varies with same object.

Refer to http://stackoverflow.com/questions/14... for finding edges of arc, organize coordinates forming arc in a vector either clock wise or anticlockwise order.. find mid value of vector, which would be the center of arc.

Looks promising, will try and update you the result of Shape context in OpenCV 3.0.

Yeah, number of points chosen plays very important role in matching. But for now I am using all the descriptor points (say, 625, 535 points for few object). I find minimum of two contours and match those points by finding distance between them.

Hey algorithm enthusiast,

Here is an implementation I am trying to make it work for shape matching that is invariant to rotation, translation and scaling as per theory.

Reference Links

• http://www.isy.liu.se/cvl/edu/TSBB08/...
• http://homepages.inf.ed.ac.uk/rbf/CVo...
• http://users.monash.edu.au/~dengs/res...
• http://users.monash.edu.au/~dengs/res...
• https://homepages.cae.wisc.edu/~ece53...
• http://www.codeproject.com/Articles/1... (ref sample application in spatial domain)

What is the process that I tried

• Get RGB input image, convert to gray scale

• Guassian Blur the image with 3x3 kernel and perform thresholding

• Do some morphological operation to close the edges and find external contour of this thresholded image

• Find the largest contour that forms the boundary of desired object. The returned contour points are in counter clock wise direction, so reverse it to make it clockwise order. (As explained in few papers)

• Perform differential coding (current coordinate - next coordinate) for contour coordinates and save it as vector of new points. Convert this to two planes (one for x and another for y).

• Perform DFT for these differential encoded contour points. Calculate Magnitude from result of DFT, discard DC coefficient and normalize other Fourier coefficients with first harmonic (this is done for scale invariancy).

• Perform Euclidean distance for template image and input test image.

Stuck up with

The Euclidean distance calculated is coming same/closer for any test sample I give. Not sure where I do mistake. I do not find any sample implementation for frequency domain shape matching.

Please help in understanding any mistakes that is being done.

Sample image in recognition

EDIT:

I am trying to perform rotation, scale and translation invariant shape matching using Fourier descriptor as proved in technical papers. I am making use of OpenCV 2.4.9 on Windows.

Could you elaborate your task and edit you question? Do you want to count number of faces in a scene? or Do you want to count occurrence of your face from a video of specific duration? Be detail while posting question in order to provide better replies.