Can anyone help me out in coding of hand detection ?

I am trying to code for REGION EXTRACTION as mentioned in statement on page # 3 of research paper attached. But do not know from how to do that.

Answer 1

[Image Analyst]

See my Image Segmentation Tutorial for an example of how to do image segmentation (what you call REGION EXTRACTION): http://www.mathworks.com/matlabcentral/fileexchange/?term=authorid%3A31862

[Explorer]

I have tried to make a code of hand detection (attached) with the help of your mentioned code(image segmention) but there is a little problem in it which hand and face are labeled by same color.

Output:

[Explorer]

@ Image Analyst

Can you please help me?

[Image Analyst]

The hand and face aren't labeled with the same color. They are not labeled at all . They are background . Invert the image if you want them to be the foreground and get labeled. From your binary image, what would you like as the output? Which blob(s)?

[Explorer]

Okay. How can I label hand and face. I want to detect hand.

[Image Analyst]

Invert, hole fill, label, take smallest blob (or blob closest in area to the expected size of a hand). You can use bwareaopen() to get rid of blobs smaller than a certain size. You can also get the N largest or smallest blobs with my attached function. For example, get the 2 largest, then the 1 largest and subtract the images to get only the second largest.

[romaisa sabih]

sir, i am doing real time hand detection but the problem i am facing is i am not able to isolate the hand from the background/environment.. i have tried background subtraction , edge detection but didn't get any fruitful result. please help

[Explorer]

1. Capture image using webcam of laptop.

2. Detect skin (You will get hand and face)

3. Make hand and face image binary

4. detect small blob to detect hand as big one will be face.

[Image Analyst]

You might want to look at motion if you have a video. The hand most likely will move but the background won't. So any blobs you detect that are the same from image to image are most likely background, not a hand.