Pattern Matching - find the fingers by recognizing the hand gesture PNG file.

[247daydreamer]

I am trying to read a hand gesture PNG image and guess the fingers count, finger name(index, thumb, ...) and which hand is it. I don't have IMAQ module, so I cannot use it.

As a first step, I am trying to detect the palm of image. Always the palm will be visible in input image. 

There are resources in internet with IMAQ and external libraries only. Unfortunately I don't have access to external library or IMAQ. 

Any support in accomplishing this task would be greatly appreciated. Looking forward for hints and suggestions. Thanks in advance. 
 

[ederzer]

I think it would be easier to use TensorFlow or any framework for deep learning; and if you want to do it with labview, you could execute python scripts on it.

[alexderjuengere]

---

@ederzer wrote:

I think it would be easier to use TensorFlow or any framework for deep learning; and if you want to do it with labview, you could execute python scripts on it.

---

The LabView  Python Node does support Python 3.6, which is supported by recent Versions  Tensorflow

The TF Object Detection API even suggest 3.6 explicitly.

Keep in mind that Tensorflow is often 64 bit only, so you need LabView 64 Bit to run your Python script

[247daydreamer1]

Appreciate your response @ederzer and @alexderjuengere. Unfortunately, I am not supposed to use any external scripts. As I mentioned in the post, no python script or IMAQ should be used. The purpose of this task is to attain LabVIEW basics knowledge. 

Thinking in a very basic way, I am trying to read the coordinates of the image to guess hand and fingers. I'd like to request you all to share resources for image traversing with basic features available in LabVIEW, that'll be really helpful. Thanks, Cheers!  

[alexderjuengere]

---

@247daydreamer wrote:

Always the palm will be visible in input image. 

---

are the input pictures always the same size`?

are those images captured by the same camera, in color or in grayscale?

how many png  images in total do you want to classify?

I could assume, that there exist at least 5 classes (fingers, gestures,...), but how many classes do exist?

show us an example image

is this some kind of homework assignment?

[ederzer]

If you are not able to use any library and you only need to use your own code, I would suggest to work with 2D arrays and for loops, you could do a program that normalizes the picture into  a 0-1 matrix, depending of the color or greyscale used in the hand picture; and after that make an image processing again with arrays that you would match the normalized 1s;

Something similar to this, and then match the patter to your expected ouputs

[alexderjuengere]

---

@ederzer wrote:

If you are not able to use any library and you only need to use your own code, I would suggest to work with 2D arrays and for loops, you could do a program that normalizes the picture into  a 0-1 matrix, depending of the color or greyscale used in the hand picture; and after that make an image processing again with arrays that you would match the normalized 1s;

like taking the sum in a region of interest...

picture from https://www.hindawi.com/journals/tswj/2014/267872/