what is the best methodology to detect playing cards?

i am trying to detect the suit and value of a playing card, OCR doesn't work for this problem, however i suspect i should use the image labeller app within matlab to label the ground truth data for each card. Any ideas of where to go from here?

[Walter Roberson]

I think OCR would work for this, if you use the facility to train additional fonts.

[philip curran]

this seems to be the most accurate method of detecting the value of each card (by creating my own language in the OCR trainer) compared to other solutions given in the responses to this post.

[Image Analyst]

I think using Deep Learning CNN clasification would be another option to consider if you have lots of different card face styles that you need it to work on.

Answer 1

[yanqi liu]

yes,sir，may be use hu，lbp or other feature to classify，or just use cnn to train data

[philip curran]

please can you explain what hu and lbp are? I would like to avoid using cnn to train the data as i have a relatively low sample size of my data.

[yanqi liu]

yes，sir，for example，we get lbp feature vector，such as

clc; clear all; close all;
im = rgb2gray(imread('football.jpg'));
v1 = extractLBPFeatures(im,'Upright',false);
im = imread('rice.png');
v2 = extractLBPFeatures(im,'Upright',false);

then use them to svm or bp to make classify model

if data not enough，may be use imageDataAugmenter to extend

Answer 2

[Image Analyst]

You could use normalized cross correlation. See attached demo.