Hi! One simple problem. I've got face pictures (different resolution etc) and I want to find an eye (retina exactly) and then isolate them from this picture. My friend told me that eye is the most darker area on picture (but when I find the most darker area - pixel by pixel) i locate hear, nose or other thinks. How can i correct it? Thanks for help. Rafal
Sure the center of the eye may be dark.... but suppose you have a pic of a woman in a polka-dot dress?....game over. What if the pic suffers with red-eye Feature recognition is about relativity. Trying to find dark circles could lead to any kind of matches. However, looking for two dark circles a little space apart would narrow your search down. Include other features of the face to narrow the search down further.Searching for the retina as a feature inside the eye would help even more.Looking for a dark circle inside a white ellipse. What exactly are you trying to achieve?....there may be a much simpler way to go about it.
I'm fairly sure there would be no accurate way to do this via PHP. This would take some very advanced coding in a different language such as assembly or C++, of course, that's my assumtions. As far as I know, no one has manipulated such a thing in PHP.
If you can do it in C or ASM, there's no reason you couldn't do it in php. I invite you to tell me why you couldn't. As an aside, ideas_man has some good suggestions, but even at that you're probably only talking an 80% success rate.
Bad assumption. It can be done in PHP (or Java or LISP or MUMPS or COBOL...), but not sure if I would want to. Is this something you need to do in real-time? Or is it a batch process?
Are the pictures in some sort of standard, like size and background colors or anything that could serve as a pattern to provide a proximity of the person eyes? It would be much easier to evaluate a small area than the entire picture.
The sad truth is that there are hundreds of scenarios that would make your project difficult. Like ideas_man said, there may be other objects in the picture that "match" for an eye. Plus you have to take into account lighting conditions, angles, colors, if the eye is looking straight ahead or away, etc. At it's most basic, you'll have to think in vectors and shapes. My understanding of the kind of stuff intended to ID people in airports and such is that the computer has formulas for the basic shapes of features. Such as an arc for an eyebrow or an ellipse for the eye. So if the program finds a circle (iris) within an ellipse (eyeball) beneath an arc (eyebrow), it is likely it found an eye. The more features it finds (nose, edge of the face, hairline, etc.) the more exact it would be.