Finding the Most Appropriate Image Standard [Image Recognition]
We have some sample images.
And we have an input image set. Each input image is one example after a combination of the following things.
1) Rotation
2) Scaling
3) The cut part of it
4) Adding noise
5) Using a filter of any color
This is a guarantee that the person can recognize the lightness of the image.
I need a simple but effective algorithm for recognition, from which one of the basic examples we get an input image.
I am writing in C # and Java
source to share
I don't think there is one simple algorithm that will allow you to recognize images under all the conditions you mentioned.
One way that can cover most is the Fourier transform , but this cannot be described as simply anywhere in the imagination, and will involve some pretty heavy mathematical concepts.
You may find it helpful to look in the Digital Signal Processing field , which includes image processing, since these are just 2D signals.
EDIT . Obviously, the problem is limited to recognizing MONEY (notes and coins), so the first search problem is to avoid websites that mention money as a result of using their image recognition product, and not as a source of images.
Anyway, I found more useful hits while searching for “ Currency Image Recognition .” Including some that mention hidden Markov patterns (whatever that means) This might be the algorithm you are looking for.
The problem is simplified by the small set of target images, but complicated by the need to detect forgeries.
I still don't think there is a "simple agorithm" for this job. Good luck with your search.
source to share
There is good research in the field of computer vision. One of the problems that is being addressed is identifying the object regardless of zoom changes, added noise and distortions, since the photo was clicked from a different view. I did a little assignment over two years ago as part of a computer vision course. There is a transformation called the scale invariant transformation functionwith which you can extract various functions for a corner point. Corner points are those that are different from all neighboring pixels. As you can see, if the photo was clicked from two different views, some of the edges may disappear and appear as something else, but the corners remain much the same. These transformations explain how a vector of 128 objects can be extracted for all corner points and tells you how to use this vector function to find out the similarity between two images. Here in your case, you can extract these functions for one of all the currency notes you have and check for these corner points in the test image you should be testing.
Since this transform is resistant to rotation, scaling, cropping, adding noise and filtering color, I think this is the best I can offer you. You can check this demo for a better understanding of what I have explained.
source to share
You will need to build a set of functions that calculate the probability of a particular transformation between two images f (A, B). A number of transformations have previously been suggested as answers, eg. Fourier. You probably won't be able to compute the probability of multiple transformations in one go fgh (A, B) with any reliability. So you have to compute the probability that each transform was independently applied f (A, B) g (A, B) h (A, B) and those with P above the threshold are the solution.
If order is important i.e. you need to know that f (A, B), then g (f, B), then h (g, B) was executed, then you would need to adopt a state-based probabilistic structure such as hidden Markov models or Bayesian network ( well, that's a generalization of HMM) to simulate the probability of moving between states. See BNT Toolkit for Matlab ( http://people.cs.ubc.ca/~murphyk/Software/BNT/bnt.html ) for more details on these or any good modern AI book.
source to share