Delete Background / Scene Image

Question

I captured the image in the first part, but then I would like to extract everything from the background that I captured on the right side (the one that does not have a hand), then to extract the hand.

My problem right now is I cannot extract the background. I tried the scan method of each pixel and see if it matches. If this happens, it will then turn the pixel into black. However, the plan failed. Is there any other way to solve this problem?

Later, I would use a neural network to determine if it is a hand or not.

Answer 1

To achieve this, you can use the BitBucket histogram method. Here is an example of his work. Image Comparison - Fast Algorithm

just save unmatched pixels, you have a hand image.

Answer 2

Try iplab to apply many filters to it. I used to have good results to remove contextual noise from the image.

EDIT: It uses the Aforge Framework .

Answer 3

I completed the task:

The first thing I did was extract skin color. (May use AForge AdaptiveSkinColor) Secondly, if it matches skin color, make it white or black.

Finally (I have a problem with this part). Recognize the form, I want to ask if anyone has any other method (better than mine) for this task.

My method: 1) Extract blob ( http://geekblog.nl/entry/24 : I used the method of these persons) and put it in an array 2) Then take one bitmap from the array and cut it into many parts. Like:

3) Check if the (fragmented part with the image) is black or white, and assign it to a double array. [If black assigns 0.0 or 1.0] 4) Use the hopfield neural network and see if this image is really what we want.

I know that my method is not so effective, but this is the method that I wrote earlier. Therefore, I ask another method to execute. (Any recommendations?)

Answer 4

A number of photo editing tools ( Topaz ReMask is a good example) will extract elements from a given image. In such programs, the user quickly draws a diagram of the object of interest. To make this task easy for the user, a rather "thick" brush is used. The program now knows that everything inside the drawn form is “subject of interest”, while everything else is “not of interest”. It remains only to deal with pixels in an indefinite boundary region. I expect the program to use a fairly simple classification process, such as linear discriminant analysis, to predict which pixels belong to the class.

Answer 5

You can also try the method that I suggested here:
HLSL Shader for Background Image Subtraction