It will be help full if you upload some sample images

It depends on the quality of images, the position of the camera and the tiles. It would be best if you can upload few images. 

Hello Mohamed

I believe you can sample the surface into blocks, then find the Principal Component Analysis (PCA) for each undeformed block and the PCA for a test block.

Then a supervised leaning technique such as back propagation neural network (BPNN) or Support Vector Machines (SVM) can be trained using the extracted PCAs features. An information fusion technique at the decision level can be used to decide if the whole surface is deformed or not.

Another approach is to match the Histogram of gradients(HOG) or SIFT features of an undeformed surface (template) with the test surface.

I hope this will solve the problem. 

best regards

Khaled

Hope texture analysis with different techniques(wavelets, PCA, etc ) along with different classifiers will be helpful. Feature extraction like color and pattern will be also useful.

PCA will be helpful.

Could you upload some sample images of good tiles and of defects?