Coin detection to make coin matching system

edit

I'm developing an matching system in java using opencv the system accept an image with one object which is coin in order to determine the value of the currency.

What I did : 1-Convert image to gray level. 2-Use edge detection in the gray image the output of this step is used as input to the next step. 3-Use SIFT to find the feature descriptors. 4-Use SVM by storing the SIFT descriptors to predict the value of image.

BUT this algorithm does not give good result. 1-what can I do in order to enhance this algorithm, 2- what is the best SVM params that is suitable for this case . 3-what is the appropriate edge detection for this purpose.

Any suggestion please. Note that in the number of coins types are five.