Training neural networks based on images

     I trained neural networks with images as inputs. I was not sure how to do that. I did like this

a=imread('A.bmp'); %I have these image files

b=imread('B.bmp');

a=a(:); %To convert this into one column matrix

b=b(:);

x=[a b];

x1=double(x); %Input to the neural network

t=eye(2); %Target and there are two classes 'a' and 'b'

After this I trained the neural network(pattern recognition) and then it got trained and I stored the network in 'net'.

After this I tried this code to check my neural network

input=a; output=sim(net,input);

I got this error

Error in ==> network.sim>simargs at 236 switch class(P)

??? Output argument "Pi" (and maybe others) not assigned during call to "D:\MATLAB\toolbox\nnet\nnet\@network\sim.m>simargs".

Error in ==> network.sim at 173 case 2, [err,X,Xi,Ai,T,Q,TS,matrixForm] = simargs(net,X);

What is the problem? Is there any other way to solve my question which is training neural networks based on images.

Basically I don't know that much about neural networks in matlab so I would like to know if there is any tool in neural network that is specially used for image recognition.

NOTE: I trained neural networks without using images (just using normal sequences) and it worked perfectly. But I don't get the same for images. Plz help me.

Answer 1

[Walter Roberson]

This has been asked by several people in several years, but solid answers seem to be scarce on this.

I did find a posting in which one person discovered that things started to work for them when they changed their input data type to double. See http://www.mathworks.com/matlabcentral/newsreader/view_thread/99987

Notice that in your code, "a" is still the data type of the original image.

[i Venky]

Thanks mate. I changed that to double and it worked.

Answer 2

[padmavathi vattikonda]

how to give an image as input for neural network and how to train it in matlab