How can I to predict a image by a trained weight parameter?

[linhj18...@gmail.com]

I have train a network by mnist dataset, and generate a weight parameter "net00002.mat". Now, if I have a hand-write picture "pic1.png" of "1", how can I to build a network to predict it and get a "int" return "1" in code way?
As we know, in paper "EBLearn: Open-Source Energy-Based Learning in C++" mention as follow:
idx<double> image = load_image<double>("im.jpg");
state_idx<double> input(image);
int answer = E.infer2(input);
However, the version is changed so much. I am confused about this question. Who can help me? Thanks.